Front cover

Best Practices for DB2 on

AIX 6.1 for POWER Systems

Explains partitioning and virtualization

technologies for Power Systems

Discusses DB2 performance

optimization on System p

Covers OLTP and data

warehouse workloads

Michael Kwok
Rakesh Dash Anupama Padmanabhan
Bernard Goelen Punit Shah
Vasfi Gucer Basker Shanmugam
Rajesh K Jeyapaul Sweta Singh
Sunil Kamath Amar Thakkar
Naveen Kumar Bharatha Adriana Zubiri
ibm.com/redbooks
International Technical Support Organization

Best Practices for DB2 on AIX 6.1 for POWER

Systems

April 2010
SG24-7821-00
 Note: Before using this information and the product it supports, read the information in
“Notices” on page xix.

First Edition (April 2010)

This edition applies to DB2 Version 9.5 and Version 9.7, AIX Version 6.1, and VIOS Version 2.1.2.
© Copyright International Business Machines Corporation 2010. All rights reserved.
Note to U.S. Government Users Restricted Rights -- Use, duplication or disclosure restricted by GSA ADP
Schedule Contract with IBM Corp.
Contents

Figures ............................................................................................................... ix

Tables .......................................................................................................................... xiii

Examples .......................................................................................................... xv

Notices ............................................................................................................. xix

Trademarks ........................................................................................................ xx

Preface ............................................................................................................. xxi

The team who wrote this book ..........................................................................xxii
Now you can become a published author, too! ................................................xxvi
Comments welcome ....................................................................................... xxvii
Stay connected to IBM Redbooks................................................................... xxvii

Chapter 1. Introduction .................................................................................... 1

1.1 Introduction to Power Systems ..................................................................... 3
1.1.1 POWER Architecture 3
1.2 Introduction to virtualization .......................................................................... 6
1.2.1 Why virtualize servers? 8
1.2.2 Virtualization benefits 8
1.3 Introduction to AIX 6.1 10
1.3.1 IBM AIX V6.1 11
1.4 Introduction to DB2 15
1.4.1 Autonomics 15
1.4.2 pureXML 17
1.4.3 Automatic storage 17
1.4.4 Performance 18
1.4.5 Reliability and scalability 19
1.4.6 Multi-vendor SQL and API support and migration to DB2 .................. 20
1.4.7 DB2 in virtualization environment ....................................................... 20
1.5 Introduction to PowerVM virtualization ....................................................... 21
1.5.1 Logical partition type ........................................................................... 25
1.5.2 Processor virtualization ...................................................................... 26
1.5.3 Memory virtualization .......................................................................... 30
1.5.4 I/O virtualization .................................................................................. 32
1.5.5 Useful links ......................................................................................... 34

Chapter 2. AIX configuration ......................................................................... 35

© Copyright IBM Corp. 2010. All rights reserved. iii
2.1 AIX configuration and tuning for DB2.......................................................... 36
2.1.1 Virtual memory considerations ........................................................... 36
2.1.2 AIX vmo command ............................................................................. 38
2.1.3 VMM considerations for DB2 .............................................................. 40
2.1.4 Large page considerations ................................................................. 46
2.1.5 Paging space considerations for DB2 ................................................ 49
2.1.6 Network tunable considerations ......................................................... 51
2.1.7 Input and output tunable considerations ............................................. 63
2.1.8 Scheduler tunable considerations ...................................................... 69
2.1.9 DB2 Groups, users, and password configuration on AIX ................... 70
2.1.10 DB2 user ID resource limits (ulimits) ................................................ 70
2.1.11 File system mount point permissions................................................ 70
2.1.12 File system mount options ................................................................ 71
2.1.13 Maximum number of AIX processes allowed per user ..................... 71
2.2 DB2 registry variables 72
2.2.1 DB2_Resource_policy 73
2.2.2 DB2 memory registry variables .......................................................... 75
2.2.3 DB2 Communications registry variables............................................. 77
2.2.4 DB2 Database Manager Configuration (DBM) parameters ................ 80
2.2.5 DB2 Database Configuration (DB) parameters .................................. 80
2.3 Configurational differences: AIX 5.3 versus AIX 6.1 ................................... 80

Chapter 3. Storage layout .............................................................................. 85

3.1 Introduction 86
3.2 DB2 storage design 87
3.2.1 Automatic storage 88
3.2.2 Tablespace design 92
3.2.3 Other considerations for DW environments ........................................ 98
3.3 Storage hardware 100
3.3.1 Introduction to the IBM Storage DS5000 Storage Server ................. 100
3.3.2 Physical components considerations ............................................... 101
3.3.3 Storage structure: RAID levels ......................................................... 104
3.3.4 Logical drives (LUN) and controller ownership ................................. 114
3.4 Tuning storage on AIX 115
3.4.1 Multipath driver 115
3.4.2 hdisk tuning 116
3.4.3 Fibre Channel adapters configuration .............................................. 118
3.5 LVM configuration: Volume groups and logical volumes .......................... 119
3.5.1 Creating and configuring VGs .......................................................... 120
3.5.2 Creating and configuring logical volumes ......................................... 125
3.6 File systems 126
3.7 Conclusion 128
iv Best Practices for DB2 on AIX 6.1 for POWER Systems
Chapter 4. Monitoring ................................................................................... 129
4.1 Understanding the system ........................................................................ 131
4.2 Benchmarking 131
4.3 Determine the possible causes................................................................. 132
4.3.1 Poor application design 132
4.3.2 Poor system and database design ................................................... 132
4.3.3 System resource shortages .............................................................. 132
4.4 Planning monitoring and tuning ................................................................ 133
4.5 Monitoring tools for DB2 ........................................................................... 134
4.5.1 Point-in-time monitoring tools ........................................................... 134
4.5.2 Traces 134
4.5.3 Snapshots 135
4.5.4 db2top 142
4.5.5 db2pd ................................................................................................ 142
4.6 Monitoring enhancements in DB2 9.7....................................................... 149
4.6.1 In-memory metrics: New monitoring infrastructure ........................... 150
4.6.2 New administrative views 160
4.7 Monitoring tools for AIX 169
4.8 Monitoring scenarios 185
4.8.1 Case 1: Alter Tablespace no file system cache ................................ 185
4.8.2 Case 2: Alleviating bottlenecks during index creation ...................... 199
4.8.3 Case 3: Alleviating I/O bottleneck ..................................................... 209
4.8.4 Case 4: Memory consumed by the AIX file system cache................ 216
4.8.5 Case 5: Disk performance bottleneck............................................... 221

Chapter 5. LPAR considerations ................................................................. 225

5.1 LPAR planning .......................................................................................... 226
5.2 LPAR profile .............................................................................................. 227
5.3 Dynamic logical partitioning ...................................................................... 229
5.4 Dedicated LPAR versus micro-partition .................................................... 231
5.4.1 LPAR throughput 232
5.4.2 LPAR scalability 232
5.4.3 Discussion 233
5.5 Uncapped micro-partition.......................................................................... 234
5.6 Micro-partition: Virtual processors ...................... 235
5.7 Multiple shared-processor pools ............................................................... 236
5.8 LPAR monitoring ................................................ 238

Chapter 6. Virtual I/O .................................................................................... 241

6.1 Virtual I/O benefits 243
6.2 VIOS features 244
6.2.1 Storage virtualization 244
6.2.2 Network virtualization 248
 Contents v
6.2.3 Live Partition Mobility 249
6.2.4 Integrated Virtualization Manager .................................................... 250
6.2.5 Shared memory paging for Active Memory Sharing ......................... 250
6.3 VIOS resilience 250
6.3.1 Dual VIOS 251
6.4 VIOS sizing 257
6.4.1 VIOS memory sizing 257
6.4.2 VIOS CPU sizing 258
6.4.3 Sizing dual VIOSs 261
6.4.4 Sizing VIOS for Live Partition Mobility .............................................. 261
6.5 VIOS best practices for DB2 ..................................................................... 261
6.5.1 Multi-path I/O 261
6.5.2 Networking 262
6.5.3 CPU settings for VIOS 264
6.5.4 SCSI queue depth 264
6.5.5 Dual VIOS 266
6.5.6 SEA threading 266
6.5.7 Best practices summary 267

Chapter 7. Live Partition Mobility ................................................................ 269

7.1 LPM overview 270
7.1.1 Active migration 270
7.1.2 Inactive migration 270
7.2 LPM........................................................................................................... 271
7.3 DB2 migration-awareness ........................................................................ 273
7.4 System planning 274
7.4.1 Managed system’s requirements ..................................................... 275
7.4.2 HMC requirements 285
7.4.3 IVM requirements 287
7.4.4 VIOS requirements 287
7.4.5 LPAR requirements 294
7.4.6 Network requirements 303
7.4.7 Storage requirements 306
7.4.8 Summary 315
7.5 Migration process and flow ....................................................................... 316
7.5.1 Migration validation 317
7.5.2 Active partition migration 320
7.5.3 Inactive partition migration 321
7.6 Mobility in action 322
7.6.1 Configuring the test environment ..................................................... 323
7.6.2 Performing the LPM 325
7.6.3 What happens during the partition migration phase? ....................... 331
7.6.4 Post migration observations ............................................................. 335
vi Best Practices for DB2 on AIX 6.1 for POWER Systems
7.7 LPM: Summary of actions ......................................................................... 354
7.7.1 LPM: Environment configuration ...................................................... 355
7.7.2 LPM: Pre-migration tests 355
7.7.3 LPM: Real migration 356
7.7.4 Post migration 356
7.7.5 Problem determination 357

Chapter 8. Workload partitioning ................................................................ 359

8.1 Overview of WPAR 360
8.2 Other types of WPARs 361
8.2.1 System WPAR 362
8.2.2 Application WPAR 363
8.3 Installing and configuring DB2 on a WPAR .............................................. 364
8.3.1 Installing the DB2 copy 365
8.4 Benefits of WPAR 367
8.4.1 When to use WPARs 368
8.4.2 When WAPR might not be the best choice ...................................... 369
8.5 Live Application Mobility versus Live Partition Mobility ............................. 370

Appendix A. System health check .............................................................. 371

Why collect ongoing data and data before any major change? ....................... 372
Why perform ongoing health checks and act on these? .................................. 372
DB2 health check ............................................................................................. 373
AIX health check .............................................................................................. 376

Abbreviations and acronyms........................................................................ 379

Related publications ...................................................................................... 381

IBM Redbooks ................................................................................................. 381
Online resources .............................................................................................. 382
How to get Redbooks....................................................................................... 382
Help from IBM .................................................................................................. 383

Index................................................................................................................ 385
Contents vii
viii Best Practices for DB2 on AIX 6.1 for POWER Systems
Figures

1-1 POWER6 architecture .................................................................................. 5

1-2 Virtualization history ..................................................................................... 7
1-3 Other virtualization approaches .................................................................. 10
1-4 Administration costs ................................................................................... 15
1-5 DB2 industry benchmarks .......................................................................... 18
1-6 DB2 pureScale ........................................................................................... 19
1-7 PowerVM Editions ...................................................................................... 23
1-8 PowerVM virtualization architecture ........................................................... 24
1-9 Profile chart................................................................................................. 26
1-10 Processors terminologies ......................................................................... 27
1-11 Weight factor calculation .......................................................................... 29
1-12 Dedicated versus shared memory environment ....................................... 31
1-13 SEA versus IVE use ................................................................................. 32
2-1 Computational and non-computational memory occupancy ...................... 36
2-2 The file system cache consumption as pointed out by numperm............... 37
2-3 Page stealing based on minfree and maxfree values ................................ 38
2-4 Output of vmo -L ......................................................................................... 39
2-5 Enabling “jumbo frames” using smitty ........................................................ 59
3-1 Typical OLTP storage layout ...................................................................... 86
3-2 Typical Data Warehouse storage layout .................................................... 87
3-3 How does a table space grow? .................................................................. 91
3-4 Storage grows with uniform capacity storage paths and rebalance ........... 92
3-5 Directly connected hosts to the DS5000 storage server .......................... 102
3-6 Hot spare coverage with alternating loops ............................................... 104
3-7 RAID 0 ...................................................................................................... 105
3-8 RAID 1 ...................................................................................................... 106
3-9 RAID 5 ...................................................................................................... 107
3-10 RAID 6 .................................................................................................... 108
3-11 RAID 10 .................................................................................................. 109
3-12 Enclosure loss protection ....................................................................... 113
4-1 Components involved ............................................................................... 131
4-2 How metrics are gathered ........................................................................ 151
4-3 Activity level monitoring ............................................................................ 155
4-4 Database object monitoring ...................................................................... 159
4-5 Identifying the resource bottleneck ........................................................... 161
4-6 Wait time versus processing time with respect to overall request time .... 164
4-7 Breakup of Total Wait Time ...................................................................... 164
4-8 Breakup of processing time with component processing time metrics ..... 165
© Copyright IBM Corp. 2010. All rights reserved. ix
4-9 Other stages of DB2 processing ............................................................... 165
4-10 vmstat output .......................................................................................... 171
4-11 VMSTAT snapshot.................................................................................. 172
4-12 Snaphot from IOSTAT showing greater value than 40% to report a IO
bottleneck ................................................................................................. 173
4-13 IOSTAT header output............................................................................ 174
4-14 Snapshot from topas is used to establish that the DB2 engine is
contributing to the CPU bottleneck .......................................................... 176
4-15 SVMON................................................................................................... 178
4-16 AIX and processor details ....................................................................... 179
4-17 nmon output ............................................................................................ 180
4-18 nmon output ............................................................................................ 181
4-19 nmon output ............................................................................................ 182
4-20 Sample CPU usage pattern .................................................................... 182
4-21 Status of the active processes ................................................................ 183
4-22 Network usage chart ............................................................................... 184
4-23 nmon graph............................................................................................. 184
4-24 nmon graph............................................................................................. 185
4-25 CPU total ................................................................................................ 210
4-26 Disk%busy .............................................................................................. 211
4-27 Disk Read KB/s....................................................................................... 212
4-28 Disk write KB/s ........................................................................................ 212
4-29 The I/O activity looks balanced across the adapters .............................. 213
4-30 nmon output ............................................................................................ 218
4-31 nmon output ............................................................................................ 222
4-32 nmon chart .............................................................................................. 222
5-1 LPAR configurations ................................................................................. 226
5-2 LPAR processor minimum, desired, maximum attributes ......................... 228
5-3 LPAR memory minimum, desired, and maximum attributes .................... 229
5-4 Dynamic logical partitioning (DLPAR) menu............................................. 230
5-5 Result of the tests ..................................................................................... 231
5-6 Compare dedicated and shared-processor partitions ............................... 232
5-7 Dedicated processor and shared processor LPAR scalability .................. 233
5-8 POWER6-based server with MSPPs defined ........................................... 237
6-1 Virtual SCSI configuration......................................................................... 245
6-2 a managed system configured to use NPIV ............................................. 247
6-3 MPIO attributes ......................................................................................... 252
6-4 Network Interface Backup using dual VIOSs ............................................ 254
6-5 SEA failover using dual VIOSs ................................................................. 256
6-6 Estimated size of storage array to drive I/O versus VIOS CPU ................ 260
6-7 Link aggregation using a single VIOS ...................................................... 263
7-1 Server capabilities .................................................................................... 277
7-2 Processor compatibility mode ................................................................... 279
x Best Practices for DB2 on AIX 6.1 for POWER Systems
7-3 Dual HMC setup ....................................................................................... 283
7-4 Desktop and rack-mount HMCs ............................................................... 285
7-5 VIOS Mover Service Partition capability for LPM ..................................... 289
7-6 Huge page settings ................................................................................... 302
7-7 Sample SEA failover setup with Etherchannel and dual VIOS setup ....... 304
7-8 SAN Architecture: Example of Disk Mapping ........................................... 309
7-9 POWER6 systems capable to migrate ..................................................... 317
7-10 Same HMC management (Remote HMC support under conditions) ..... 318
7-11 VIOS configured ..................................................................................... 318
7-12 All disks must be accessible by both systems ........................................ 319
7-13 Operating System level .......................................................................... 319
7-14 Network connectivity with all partitions and HMC ................................... 320
7-15 Configuring JS43 for LPM ...................................................................... 322
7-16 DB2 environment settings ...................................................................... 323
7-17 lsconf output prior to the migration ......................................................... 326
7-18 Migration menu ....................................................................................... 327
7-19 Migration validation ................................................................................. 329
7-20 Observe the status of the migration on the destination server ............... 330
7-21 Suspension: Resumption of the LPAR during Migration ........................ 332
7-22 Partition migration in progress ................................................................ 333
7-23 lsconf output after migration occurred .................................................... 334
7-24 LPM hand over ....................................................................................... 336
7-25 Mobile partition: Total CPU..................................................................... 337
7-26 Mobile partition: Processes RunQueue .................................................. 337
7-27 VIOS1: Physical CPU versus EC ........................................................... 338
7-28 VIOS2: Physical CPU versus EC ........................................................... 338
7-29 Mobile partition: Disk use: Total (KB/second) ........................................ 339
7-30 Mobile partition: Disk transfer ................................................................. 339
7-31 Mobile partition: Disk adapter use .......................................................... 340
7-32 VIOS1: Disk adapter use ........................................................................ 340
7-33 VIOS2: Disk adapter use ........................................................................ 341
7-34 Mobile partition: Memory use ................................................................. 341
7-35 VIOS1: Memory use ............................................................................... 342
7-36 VIOS2: Memory use ............................................................................... 342
7-37 Mobile partition: Network I/O .................................................................. 343
7-38 Snapshot of a ping command ................................................................. 344
7-39 VIOS1: Network I/O ................................................................................ 345
7-40 VIOS2: Network I/O ................................................................................ 345
7-41 VIOS1: Physical CPU ............................................................................. 346
7-42 VIOS2: Physical CPU ............................................................................. 347
7-43 Mobile LPAR: Physical CPU................................................................... 347
7-44 VIOSs run queue .................................................................................... 348
7-45 Mobile partition: Run queue.................................................................... 349
Figures xi
7-46 Mobile: Disk use ..................................................................................... 349
7-47 VIOS1: Disk use ..................................................................................... 350
7-48 VIOS2: Disk use ..................................................................................... 350
7-49 VIOS1: Memory use ............................................................................... 351
7-50 VIOS2: Memory use ............................................................................... 351
7-51 Mobile partition: Memory use ................................................................. 352
7-52 VIOS1: Network I/O ................................................................................ 353
7-53 VIOS2: Network I/O ................................................................................ 353
7-54 Mobile partition: Network I/O .................................................................. 354
7-55 LPM: Environment configuration ............................................................ 355
7-56 LPM: Pre-migration tests (validation) ..................................................... 355
7-57 LPM: Real migration ............................................................................... 356
7-58 LPM: After migration ............................................................................... 356
7-59 LPM: Problem determination, part 1 of 2 ................................................ 357
7-60 LPM: Problem determination, part 2 of 2 ................................................ 358
8-1 LPAR and WPAR ..................................................................................... 360
8-2 Global environment and WPAR................................................................ 362
xii Best Practices for DB2 on AIX 6.1 for POWER Systems
Tables

1-1 AIX V6.1 new features summary ................................................................ 12

2-1 Tunables: AIX 5.3 versus AIX 6.1 .............................................................. 80
3-1 Page size versus row length:...................................................................... 93
3-2 RAID levels comparison ........................................................................... 110
6-1 Approximate CPU amount VIOS needs for 1 GB of network traffic.......... 258
7-1 List of default parameters (DB2 9.7) – plus many more ........................... 273
7-2 Managed system’s requirements ............................................................. 275
7-3 HMC requirements ................................................................................... 285
7-4 VIOS requirements ................................................................................... 287
7-5 LPAR requirements for active migration ................................................... 294
7-6 LPAR requirements for Inactive migration ................................................ 295
7-7 Operating system level needed for active partition mobility ..................... 297
7-8 Operating system level needed for inactive partition mobility .................. 297
7-9 Network requirements .............................................................................. 303
7-10 Storage requirements ............................................................................. 306
7-11 Logical partition attributes that might change after a migration .............. 316
8-1 LPAR versus WPAR ................................................................................. 361
8-2 Basic differences between Partition Mobility and Application Mobility ..... 370
A-1 Ongoing data and data before any major change .................................... 373
A-2 Ongoing health checks............................................................................. 374
A-3 AIX health check tools .............................................................................. 376
© Copyright IBM Corp. 2010. All rights reserved. xiii
xiv Best Practices for DB2 on AIX 6.1 for POWER Systems
Examples

2-1 Output of “vmo -a” ...................................................................................... 39

2-2 vmo to make the changes permanent across reboot .................................. 40
2-3 Setting the maxfree value ........................................................................... 41
2-4 svmon -G .................................................................................................... 44
2-5 lsps -a ......................................................................................................... 44
2-6 vmstat-Monitor pi/po for paging activity ...................................................... 44
2-7 vmstat -v | grep num ................................................................................... 45
2-8 svmon -Put 1 .............................................................................................. 45
2-9 vmstat output showing the memory statistics ............................................. 50
2-10 lsps -a command ........................................................................ 50
2-11 Output of no -a command ................................................................... 51
2-12 Modify and make changes permanent across reboot .............................. 52
2-13 Modification takes effect at the next reboot .............................................. 52
2-14 Verification of overflows ............................................................................ 53
2-15 ipqmaxlen ................................................................................................. 54
2-16 netstat -m , showing the memory statistics .............................................. 60
2-17 netstat -an -f inet ....................................................................................... 62
2-18 netstat -i .................................................................................................... 62
2-19 Output of the command, ioo -a, to display the current value .................... 63
2-20 To modify and make the changes permanent .......................................... 63
2-21 To modify and make the changes after reboot ......................................... 63
2-22 ps vg | egrep "aio|SIZE"............................................................................ 68
2-23 changing aio_maxservers value ............................................................... 69
2-24 Create required groups, users and set initial password ........................... 70
2-25 chuser command ...................................................................................... 70
2-26 Resource sets defined in /etc/rsets file ..................................................... 74
2-27 Resource names in db2nodes.cfg file ...................................................... 75
3-1 Estimate the OVERHEAD and TRANSFERRATE ..................................... 96
3-2 lspv command .......................................................................................... 123
3-3 lsvg command .......................................................................................... 124
4-1 Buffer Pool Data/Index Hit Ratio............................................................... 135
4-2 Buffer pool Temporary Data/Index Hit Ratio ............................................ 136
4-3 Overall Average Read Time ..................................................................... 136
4-4 Average Asynchronous/Synchronous Read Time.................................... 137
4-5 Average Write Time .................................................................................. 137
4-6 Average Asynchronous/Synchronous Write Time .................................... 137
4-7 Query SYSIBMADM.BP_READ_IO & SYSIBMADM.BP_WRITE_IO ...... 138
4-8 Average Log Write Time ........................................................................... 138
© Copyright IBM Corp. 2010. All rights reserved. xv
4-9 Page Cleaning Ratio ................................................................................. 138
4-10 db2pd –edus ........................................................................................... 143
4-11 Memory set ............................................................................................. 144
4-12 db2pd -mempools ................................................................................... 145
4-13 db2pd -tcbstat ......................................................................................... 146
4-14 db2pd – activestatements....................................................................... 147
4-15 db2pd –dynamic ..................................................................................... 148
4-16 db2pd – static ......................................................................................... 149
4-17 List the active connection ....................................................................... 152
4-18 Drill down to the UOW level.................................................................... 153
4-19 Query the MON_GET_SERVICE_SUBCLASS ...................................... 154
4-20 Determine the activity_id ........................................................................ 156
4-21 Determine which SQL statement corresponds to this activity ................ 157
4-22 What are my top 5 hottest SQL statements, sorted by Rows Read? ..... 158
4-23 Determine the tables that are generating table scan.............................. 160
4-24 Converting time spent metrics as percentage of total request time........ 167
4-25 What is the breakup of the wait times inside DB2 .................................. 167
4-26 vmstat ..................................................................................................... 168
4-27 Where is DB2 spending its time in processing ....................................... 168
4-28 VMSTAT snapshot ................................................................................. 172
4-29 List paging space .................................................................................... 174
4-30 vmstat output .......................................................................................... 186
4-31 Initial Database Manager Snapshot ....................................................... 186
4-32 Database Snapshot taken at 11/06/2009 20:59:51.556288 ................... 188
4-33 Subsequent database snapshot (11/06/2009 21:01:00.612305) ........... 192
4-34 Tablespace snapshot ............................................................................. 196
4-35 Perfcount option of a DB2 trace ............................................................. 198
4-36 Tablespace definitions ............................................................................ 199
4-37 Table spaces are defined on the following physical/logical volumes...... 200
4-38 dbm and db configuration parameters .................................................... 201
4-39 Index created .......................................................................................... 201
4-40 db2pd -edus output ................................................................................. 201
4-41 Disk use .................................................................................................. 203
4-42 Index itself is being written out to the index table space ........................ 203
4-43 db2pd -table spaces output during index creation.................................. 204
4-44 Shared sort usage using either db2mtrk ................................................ 205
4-45 Script....................................................................................................... 205
4-46 Create the same index ........................................................................... 205
4-47 db2pd -edus output ................................................................................. 206
4-48 Observing the I/O on the disks ............................................................... 207
4-49 Observing the I/O on the disks ............................................................... 207
4-50 db2pd -table spaces output during the index creation............................ 208
4-51 db2pd with the mempools output............................................................ 209
xvi Best Practices for DB2 on AIX 6.1 for POWER Systems
4-52 CPU use ................................................................................................. 210
4-53 LUNs to file systems ............................................................................... 213
4-54 DB2 table space containers and log ....................................................... 214
4-55 Tables that are active in TS_TXM table space ....................................... 214
4-56 Buffer pool Hit Ratio ............................................................................... 215
4-57 Average read time of 72 ms points to poor I/O response time ............... 215
4-58 hdisk30 use............................................................................................. 215
4-59 vmstat output .......................................................................................... 216
4-60 lsps -a ..................................................................................................... 217
4-61 vmstat -v ................................................................................................. 217
4-62 svmon output .......................................................................................... 219
4-63 lsfs output ............................................................................................... 220
4-64 Tablespace snapshot ............................................................................. 221
5-1 LPAR information using lparstat command .............................................. 238
5-2 lparstat command ..................................................................................... 239
5-3 vmstat for shared processor LPAR .......................................................... 240
7-1 Output of the lssyscfg command .............................................................. 282
7-2 Output of the command lssyscfg .............................................................. 283
8-1 WPAR command ...................................................................................... 364
8-2 Listing the characteristics of WPAR ......................................................... 365
8-3 Log in to system WPAR using “clogin” ..................................................... 366
8-4 Create groups and users .......................................................................... 366
8-5 Update the /etc/hosts file .......................................................................... 367
8-6 Create the DB2 Instance .......................................................................... 367
Examples xvii
xviii Best Practices for DB2 on AIX 6.1 for POWER Systems
Notices

This information was developed for products and services offered in the U.S.A.
IBM may not offer the products, services, or features discussed in this document in other countries. Consult
your local IBM representative for information on the products and services currently available in your area.
Any reference to an IBM product, program, or service is not intended to state or imply that only that IBM
product, program, or service may be used. Any functionally equivalent product, program, or service that
does not infringe any IBM intellectual property right may be used instead. However, it is the user's
responsibility to evaluate and verify the operation of any non-IBM product, program, or service.

IBM may have patents or pending patent applications covering subject matter described in this document.
The furnishing of this document does not give you any license to these patents. You can send license
inquiries, in writing, to:
IBM Director of Licensing, IBM Corporation, North Castle Drive, Armonk, NY 10504-1785 U.S.A.
The following paragraph does not apply to the United Kingdom or any other country where such
provisions are inconsistent with local law: INTERNATIONAL BUSINESS MACHINES CORPORATION
PROVIDES THIS PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR
IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT,
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some states do not allow disclaimer
of express or implied warranties in certain transactions, therefore, this statement may not apply to you.
This information could include technical inaccuracies or typographical errors. Changes are periodically made
to the information herein; these changes will be incorporated in new editions of the publication. IBM may
make improvements and/or changes in the product(s) and/or the program(s) described in this publication at
any time without notice.

Any references in this information to non-IBM Web sites are provided for convenience only and do not in any
manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the
materials for this IBM product and use of those Web sites is at your own risk.

IBM may use or distribute any of the information you supply in any way it believes appropriate without
incurring any obligation to you.

Information concerning non-IBM products was obtained from the suppliers of those products, their published
announcements or other publicly available sources. IBM has not tested those products and cannot confirm
the accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on
the capabilities of non-IBM products should be addressed to the suppliers of those products.

This information contains examples of data and reports used in daily business operations. To illustrate them
as completely as possible, the examples include the names of individuals, companies, brands, and products.
All of these names are fictitious and any similarity to the names and addresses used by an actual business
enterprise is entirely coincidental.

COPYRIGHT LICENSE:

This information contains sample application programs in source language, which illustrate programming
techniques on various operating platforms. You may copy, modify, and distribute these sample programs in
any form without payment to IBM, for the purposes of developing, using, marketing or distributing application
programs conforming to the application programming interface for the operating platform for which the
sample programs are written. These examples have not been thoroughly tested under all conditions. IBM,
therefore, cannot guarantee or imply reliability, serviceability, or function of these programs.
© Copyright IBM Corp. 2010. All rights reserved. xix
Trademarks
IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of International Business
Machines Corporation in the United States, other countries, or both. These and other IBM trademarked
terms are marked on their first occurrence in this information with the appropriate symbol (® or ™),
indicating US registered or common law trademarks owned by IBM at the time this information was
published. Such trademarks may also be registered or common law trademarks in other countries. A current
list of IBM trademarks is available on the Web at http://www.ibm.com/legal/copytrade.shtml

The following terms are trademarks of the International Business Machines Corporation in the United States,
other countries, or both:

Active Memory™ Micro-Partitioning™ Redpaper™

AIX 5L™ Optim™ Redbooks (logo) ®
AIX® Power Architecture® RS/6000®
BladeCenter® POWER Hypervisor™ solidDB®
Blue Gene® Power Systems™ System i5®
DB2® POWER4™ System i®
DS4000® POWER5™ System p5®
DS6000™ POWER6® System p®
DS8000® PowerHA™ System Storage™
eServer™ PowerPC® Tivoli®
GPFS™ PowerVM™ TotalStorage®
HACMP™ POWER® WebSphere®
i5/OS® pSeries® Workload Partitions Manager™
IBM® pureXML®
InfoSphere™ Redbooks®

The following terms are trademarks of other companies:

ITIL is a registered trademark, and a registered community trademark of the Office of Government
Commerce, and is registered in the U.S. Patent and Trademark Office.

Snapshot, and the NetApp logo are trademarks or registered trademarks of NetApp, Inc. in the U.S. and
other countries.
Java, and all Java-based trademarks are trademarks of Sun Microsystems, Inc. in the United States, other
countries, or both.

Windows, and the Windows logo are trademarks of Microsoft Corporation in the United States, other
countries, or both.

UNIX is a registered trademark of The Open Group in the United States and other countries.

Linux is a trademark of Linus Torvalds in the United States, other countries, or both.

Other company, product, or service names may be trademarks or service marks of others.
xx Best Practices for DB2 on AIX 6.1 for POWER Systems
Preface

This IBM® Redbooks® publication presents a best practices guide for DB2® and
InfoSphere™ Warehouse performance on a AIX® 6L with Power Systems™
virtualization environment. It covers Power hardware features such as
PowerVM™, multi-page support, Reliability, Availability, and Serviceability (RAS)
and how to best exploit them with DB2 LUW workloads for both transactional and
data warehousing systems.

The popularity and reach of DB2 and InfoSphere Warehouse has grown in recent
years. Enterprises are relying more on these products for their mission-critical
transactional and data warehousing workloads. It is critical that these products
be supported by an adequately planned infrastructure. This publication offers a
reference architecture to build a DB2 solution for transactional and data
warehousing workloads using the rich features offered by Power systems.

IBM Power Systems have been leading players in the server industry for
decades. Power Systems provide great performance while delivering reliability
and flexibility to the infrastructure.

This book presents a reference architecture to build a DB2 solution for

transactional and data warehousing workloads using the rich features offered by
Power systems. It aims to demonstrate the benefits DB2 and InfoSphere
Warehouse can derive from a Power Systems infrastructure and how Power
Systems support these products.

The book is intended as a guide for a Power Systems specialist to understand

the DB2 and InfoSphere Warehouse environment and for a DB2 and InfoSphere
Warehouse specialist to understand the facilities available for Power Systems
supporting these products.
© Copyright IBM Corp. 2010. All rights reserved. xxi
The team who wrote this book
This book was produced by a team of specialists from around the world.

Rakesh Dash is a performance engineer for the ECM

Performance Engineering team with IBM. He has been
working with IBM for more than five years and has
performance consulting and development experience for
several IBM products such as DB2, AIX, WebSphere®,
IBM Java™, IBM filenet P8. He has authored several
white papers on performance best practices and
presented the performance best practices methodology
in conferences.

Bernard Goelen is an Advisory IT Specialist in

Luxemburg with more than 15 years experience with IBM
and business partners. He works on national and
international deals, on Power Systems platforms, with
extensive experience in AIX, PowerHA™, and Power
Virtualization techniques, Live Partition Mobility. He has
a key role within IBM Global Technology Services and
has an extended role within the IBM System Sales
Implementation Services department where he acts as a
pre-sales and technical support on large deals. He
focuses on server consolidation and IT virtualization and
has field experience in virtualization solutions
implementations, PowerVM and PowerHA clusters
deployments in complex environments, performance
audits, and analysis.

Vasfi Gucer is a project leader at the ITSO in Austin,

Texas. He has more than 15 years of experience in
Systems Development and Enterprise Systems
Management. He writes extensively and teaches IBM
classes worldwide on Tivoli® software. Vasfi is also an
IBM Certified Senior IT Specialist, PMP, and ITIL®
Expert.
xxii Best Practices for DB2 on AIX 6.1 for POWER Systems
Rajesh K Jeyapaul is an AIX Development Support
Specialist in IBM India. He is specialized in investigating
the performance impact of the applications on
AIX/System p®. Currently, he is a Technical Lead for
AIX Development Support team. He has co-authored an
IBM Redbooks publication on DS8000® performance
Monitoring. His area of expertise include Power
Virtualization, AIX, and High-Availability Cluster
Multi-Processing (HACMP™). He holds a Masters
Degree in Software Systems from the University of BITS,
India and an MBA from University of MKU, India.

Sunil Kamath is a Senior Technical Staff Member and

Senior Manager in Information Management at the IBM
Toronto Software Labs. He has responsibility for
performance, benchmarking, and solutions development
for DB2 and solidDB®. Sunil works primarily on
database performance and scalability and has led many
successful world-record TPC-C and SAP benchmarks.
In addition, he has designed and implemented many
high performance DB2 database solutions for customers
world-wide. Sunil is also the DB2 kernel architect for
exploiting hardware, virtualization, operating systems,
storage, and compiler technologies in DB2.

Naveen Kumar Bharatha is an independent database

consultant. He has 13 years of IT experience in
Database, Systems, and Storage design. His areas of
expertise and responsibilities include design of single
and multi-partition databases, database administration,
virtualization and automation to provide DB2
solutions. He has performed several database and
systems migrations in complex environments. He has
special interest in creating a Linux® Live OS with
persistent DB2 Database. He has worked as Staff
Software Engineer with IBM Lab services focusing on
customer engagements, POCs involving DB2. He is an
IBM certified Database Administrator in DB2 UDB
v8.1. He holds a Bachelors degree from JNTU,
Hyderabad, India.
Preface xxiii
Michael Kwok is an Advisory Software Engineer in DB2
for Linux, UNIX®, and Windows®. He works in the DB2
performance QA team as a Data Warehouse Team
Lead. His main focus is the overall performance
experience in Data Warehouse workloads. He provides
technical support on customers' performance problems.
Michael Kwok received his Ph.D. degree in Computer
Science. His doctoral dissertation is in the area of
scalability analysis of distributed systems.

Anupama Padmanabhan works as a Technical

Consultant in the ISV Solutions Enablement Group for
the enablement of SWG family of products on IBM
eServer™ platforms. She has 10 years of experience
with IT involving leadership and consulting on financial
systems, extensive experience in full software
development life cycles, including requirements
definition, prototyping, design, coding, testing and
maintenance. She holds a Bachelor's degree in
Electronics Engineering from University of Bombay,
India.

Punit Shah has a background in both DB2 internal and

AIX, POWER®, and virtualization technologies. In his
current role, Punit is in the DB2 development
organization, engaged in DB2 kernel data protection
services and backup/restore design and development for
DB2 pureScale feature. In his previous role, Punit was
architect and leader of System p/IBM Software group
enablement team. There, he was responsible for
DB2/System p/AIX stack enablement, performance and
deep technical exploitation, ensuring that DB2 (and
other IBM software products) are using latest AIX and
System p technologies, including PowerVM virtualization
feature. He architected and implemented several joint
DB2 and System p virtualization solutions and features
in DB2. In the process, he applied for several patents in
the area, and authored around dozen white papers.
Other AIX, POWER technologies he enabled in DB2 are
decimal floating point (DFP), AIX multi-page support,
and JFS2 concurrent I/O (CIO).
xxiv Best Practices for DB2 on AIX 6.1 for POWER Systems
Basker Shanmugam works in the IBM DB2
Performance and Solution Development organization
focusing on DB2 performance, benchmarking, and
developing DB2 solutions with System p virtualization
features. He has published several TPC benchmarks
and has co-authored several papers on DB2 with
System p virtualization. Basker has worked in technical
support and performance-related areas that involve
Windows, Linux and AIX operating systems, as well as
Oracle, DB2 and server virtualization for more than 19
years.

Sweta Singh is a DB2 Performance Analyst. She works

in the DB2 Performance QA Team and has worked
extensively in analyzing and improving performance of
DB2 codebase over the past six years. She is currently
the technical lead for OLTP and focuses on improving
the overall customer experience in OLTP. She is also
actively involved in customer situations related to DB2
performance.

Amar Thakkar is the Team Lead of the DB2 Down

Systems Support team in Sydney, Australia. He has over
five years experience in diagnosing critical problems and
ensuring that clients database downtime is minimized.
Amar has extensive knowledge in the areas of migration,
memory usage, backup, recovery, and
performance-related issues. He enjoys investigating
complex issues and resolving them to ensure their
impact on business is mitigated.

Adriana Zubiri is the technical manager for the DB2

Performance Quality Assurance team in the IBM Canada
lab. She has been part of the DB2 development
organization for more than 12 years, where she has led
the customer support team, participated in world-record
TPCH Benchmarks and designed and implemented new
features in the area of join processing. She has worked
extensively with customers and specialized in the
performance area for data warehouses. She holds a
Masters of Science degree in Computing Science from
University of Alberta, Canada.
Preface xxv
 Thanks to the following people for their contributions to this project:

Guido Somers
IBM Belgium

Rimas Kalesnykas, Konwen Kuan, Heidi Scott, Malcolm Singh

IBM Canada

Abhilash J Jils, Prabhu S Padashetty

IBM India

Doreen Stein
IBM Germany

Dan Braden, Grover C. Davidson, Brian Hart, James Hoy, Bart Jacob, Pete
Jordan, Jessica Scholz, Jaya Srikrishnan, Stephen M Tee, Brian Twichell, Scott
Vetter
IBM USA

David Hepple
IBM Ireland

The team would like to express special thanks to Robert Taraba from IBM USA
for providing the project resources and his support throughout the project.

Now you can become a published author, too!

Here's an opportunity to spotlight your skills, grow your career, and become a
published author - all at the same time! Join an ITSO residency project and help
write a book in your area of expertise, while honing your experience using
leading-edge technologies. Your efforts will help to increase product acceptance
and customer satisfaction, as you expand your network of technical contacts and
relationships. Residencies run from two to six weeks in length, and you can
participate either in person or as a remote resident working from your home
base.

Find out more about the residency program, browse the residency index, and
apply online at:
ibm.com/redbooks/residencies.html
xxvi Best Practices for DB2 on AIX 6.1 for POWER Systems
Comments welcome
Your comments are important to us!

We want our books to be as helpful as possible. Send us your comments about

this book or other IBM Redbooks publications in one of the following ways:
› Use the online Contact us review Redbooks form found at:
ibm.com/redbooks
› Send your comments in an e-mail to:
[email protected]
› Mail your comments to:
IBM Corporation, International Technical Support Organization
Dept. HYTD Mail Station P099
2455 South Road
Poughkeepsie, NY 12601-5400

Stay connected to IBM Redbooks

› Find us on Facebook:
http://www.facebook.com/pages/IBM-Redbooks/178023492563?ref=ts
› Follow us on twitter:
http://twitter.com/ibmredbooks
› Look for us on LinkedIn:
http://www.linkedin.com/groups?home=&gid=2130806
› Explore new Redbooks publications, residencies, and workshops with the
IBM Redbooks weekly newsletter:
https://www.redbooks.ibm.com/Redbooks.nsf/subscribe?OpenForm
› Stay current on recent Redbooks publications with RSS Feeds:
http://www.redbooks.ibm.com/rss.html
Preface xxvii
xxviii Best Practices for DB2 on AIX 6.1 for POWER Systems
 1

Chapter 1. Introduction
This IBM Redbooks publication describes the best practices to configure a DB2
database server for data warehouse or OLTP environment on a Power System
architecture-based system running AIX 6.1.

Running the DB2 product on a Power System using the right blend of
virtualization features and meeting your business goals is a real challenge that
allows you to reduce various costs (such as power, floor space, and
administration) by consolidating your servers. Sharing your system resources,
dynamically resource allocation without rebooting, and processor utilization are
key features that help you optimize the performances of your DB2 product.

The target audience of this book is database administrators and system

administrators who want to optimize the performance of their DB2 database
server for data warehouse or OLTP environments on AIX V6.1. It outlines those
particular parameters you can set to follow the best practices when configuring a
DB2 environment on AIX.

Moreover, this book is designed to help you understand the major differences
that exist when running your database on AIX 5L™ compared to running your
database on AIX 6.1. This new version of AIX introduces many new features,
including workload partitions, advanced security, continuous availability, and
managing and monitoring enhancements.
© Copyright IBM Corp. 2010. All rights reserved. 1
DB2 9.7 also offers a wide range of new features, including autonomics, data
compression, pureXML®, automatic storage, performance optimization and
security, reliability, and scalability. Most of these features are covered in this
book.

This chapter makes an introduction to the concepts that are detailed throughout
the book and contains the following sections:
› “Introduction to Power Systems” on page 3
› “Introduction to virtualization” on page 6
› “Introduction to AIX 6.1” on page 10
› “Introduction to DB2” on page 15
› “Introduction to PowerVM virtualization” on page 21
2 Best Practices for DB2 on AIX 6.1 for POWER Systems
1.1 Introduction to Power Systems
When setting up a Power Systems environment, there are many options that you
must consider and plan for properly to achieve the optimal performance goal.
Especially for a database server, the database system is the primary application
in the system. Database configuration considerations must also be incorporated
in the Power Systems planing setup. However, the Power Systems setup tasks
are usually performed by the system support team without the database
administrator’s (DBA) involvement, or the DBA might not have the expertise in
this area. The following list details the main points you go through when
configuring your logical partitions on a Power System for your database server:
› Setting up and configuring logical partitions
– Choosing a partition type
– Configuring physical and virtual processors
– Configuring the memory
› Considerations for disk storage and choosing from various options
› Considerations for the network and choosing from various options Setting up
the Virtual I/O Server (VIOS).

1.1.1 POWER Architecture

POWER Architecture technology is an instruction-set architecture (ISA) that
spans applications from consumer electronics (such as gaming consoles and
set-top boxes), Internet-enabling technologies (such as routers and switches) to
the world's fastest supercomputer, and even Spirit and Opportunity, the Mars
rover supercomputers. The first computers from IBM to incorporate the POWER
Architecture were called the “RISC System/6000" or RS/6000®.

Note: POWER stands for Performance Optimization With Enhanced RISC.

Power Architecture® refers both to POWER processors used in IBM servers and
to PowerPC® processors, which can be found in a variety of embedded systems
and desktops.

Note: PowerPC stands for POWER Performance Computing.

Chapter 1. Introduction 3
Power Architecture encompasses PowerPC, POWER4™, POWER5™ and
POWER6® processors.
› Power Architecture drives the world's highest-performance supercomputers,
including the #1 Blue Gene®.
› Power processor technology is at the heart of the Power System i® and
Power System p, IBM servers.
› The most robust storage solutions, including the IBM TotalStorage®
DS6000™ and IBM TotalStorage DS8000, incorporate Power Architecture
processors.
› Power Architecture is the leading processing platform for wireless
infrastructure, enterprise routing and switching, and wireline
telecommunications.
› VMX (vector multimedia extension) implementations of PowerPC that are key
to the gaming industry.
› Reliability and performance have made Power Architecture chips a favorite
choice for the aerospace industry.
› The POWER processor-based systems offer Virtualization Engine capabilities
such as IBM Micro-Partitioning™ technology, which enables partitions as
small as 1/10th of a processor and automated, dynamic resource
management between partitions.

As of the writing of this book, the latest in the series of POWER processor is
POWER6 processor. The POWER6 chip is a dual-core processor that runs at
speeds between 4 GHz and 5 GHz depending on the type and model of the
system. Figure 1-1 on page 5 shows the POWER6 architecture.
4 Best Practices for DB2 on AIX 6.1 for POWER Systems
 1/2
Clock

8B x2 rd
32 MB L3 Control 4B read ½ or 1/3 or ¼
I/O Ctlr
L3 Directory 4B write Clock
8B x2 wr
4B/8B x2
Fabric Switch 2x Memory Ctlr Interface
4B/8B x2

8B x2 4B x2 400, 533, 667 (800) MHz DDR2

read write
1/2 (1066 MHz DDR3)
2B/8B x2 2B/8B x2 2B/8B x2
Clock
Memory

Figure 1-1 POWER6 architecture

The following are the characteristics of the POWER6 architecture:

› The POWER6 processor implements the 64-bit IBM Power Architecture
technology.
› The POWER6 core is a ultra-high-frequency design that is optimized for
performance for the server market.
› POWER6 processor core includes two accelerators for increasing the
performance of specific workloads
– Vector Multimedia extension (VMX) provides vector acceleration of graphic
and scientific workloads.
– Decimal Floating point unit (DFU) provides acceleration of commercial
workloads, more specifically, financial transactions.

Note: The IBM POWER6 processor is the first commercial hardware

implementation of the IEEE 754R Floating-Point Arithmetic Standard that
defines decimal floating-point formats, on which substantial amount of work is
done to improve the performance of commercial workloads.
Chapter 1. Introduction 5
 › The POWER6 processor implements a two-thread SMT for each core.
› The POWER6 processor design reduces the effective latency of main
memory storage references with a substantial set of design features, such as
three levels of cache, SMT, and Load Lookahead (LLA). It enhances data
stream prefetching.
› The POWER6 processor-based systems introduce new features such as
Processor Instruction Retry, which allows the system to recover transparently
without an impact on a partition using the core.
› Dynamic Processor Deallocation enables automatic deconfiguration of an
error-prone processor core before it causes an unrecoverable system error.
› POWER6 processor-based systems have a number of protection schemes,
such as hardware scrubbing and ECC ,which are designed to prevent,
protect, or limit the effect of errors in main memory.
› POWER6 processor-based systems are designed with cache protection
mechanisms.
› PCI Error Recovery been extended to PCIe buses as well in POWER6.

In summary, with its high-frequency core architecture, enhanced SMT

capabilities, balanced system throughput, and scalability and extensions, the
POWER6 microprocessor provides a higher level of performance. Improvements
in functionality, RAS (Reliability, Availability, Serviceability), and power
management have resulted in valuable new characteristics of POWER6
processor-based systems.

1.2 Introduction to virtualization

Nowadays, IT Managers are concerned with the total cost of ownership (TCO) to
run an IT environment (such as the hardware costs, floor space the hardware
takes in a computer room, real estate prices, the power consumption, the heat
produced). Moreover, reliability, availability, and serviceability are other aspects
that everyone is concerned with.

Virtualization can help you consolidate your multiple systems, running multiple
environments and applications on a single system, all into separated
environments, and fully secured, as though you were running on an isolated
single standalone servers.

Virtualization is not new to IBM. IBM developed virtualization techniques in the

1960s. You can see the complete story on virtualization in Figure 1-2 on page 7.
6 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 1-2 Virtualization history

Virtualization is no longer an option for most data centers. Recent developments

and shifts in the marketplace, coupled with high administration cost and energy
cost, have elevated virtualization to a must-have deployment strategy.
Virtualization offers functions and benefits that allow you to share a physical
resource, consolidate multiple separated resources, and emulate hardware
resources.

Sharing a resource to create multiple images (such as LPARs, virtual disks,

virtual LANs [VLAN]) offers you the benefit of global resource use, which
increases flexibility, and eases the management of your infrastructure, all of
which takes place in a complete isolated environment.

Aggregation allows you to pool multiple separated distributed resources and

make them appear as one single resource to the client. Virtual disk, SVC SAN
Volume Controller, and RAID technology are resources that ease management,
help to protect your investment, and offer a high level of scalability.

The emulation function allows you to make use of objects (such as virtual tape,
iSCSI) that seem to be real (although no physical resource exists). It helps in the
compatibility and interoperability mode of your resources and the flexibility of
your environment.

Note: DB2 has been engineered to integrate well into virtualized

environments.
Chapter 1. Introduction 7
1.2.1 Why virtualize servers?
Virtualization helps you to create a virtual machine easily, allowing rapid
deployment of your new application using logical partitions. The overall cost of
your hardware decreases by consolidating your smaller systems onto one bigger
one.

When you increase the energy efficiency of your infrastructure by decreasing

your power consumption, you meet environmental goals. Moving logical
partitions from one server to another enables one of the two servers to power off
and so save energy.

Virtualization also offers the flexibility to maximize capacity, allowing you to move
your resources with DLPAR operations or LPM (Live Partition Mobility)
capabilities.

What virtualization offers you is a consolidated infrastructure on a larger server

on which you can maximize your resources, host numerous virtual machines that
you can build with just a click, dynamically provision, and host. Moreover, it gives
you the illusion that you have more hardware resources than really exist on the
physical machine. It offers you totally isolated virtual machines run-time
environments, allowing a mix of operating systems with greater manageability
and efficiency through balancing your system resources to virtual machines that
need resources the most. You can keep your SLAs (service level agreements)
and increase the availability and security of your servers, lower the cost of
availability, and protect your investment.

Note: Carefully consider planning before moving to a virtualized environment.

Communicate with your development team and DBAs so that all work in
concert with IT to make the virtualization a success. Do not make virtualization
more complex than necessary.

1.2.2 Virtualization benefits

Virtualization helps you reduce your hardware costs. When the physical
resources use goes higher, the footprints reduce drastically. Moreover, it
improves flexibility and responsiveness because your virtual resources can be
dynamically adjusted.

Virtualization is the key enabler of on-demand operating environments. Due to

having fewer physical servers to manage, many management tasks become
much easier. Also, the management cost is reduced.
8 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Note: Application software licensing costs can decrease when virtualizing
your environment on a consolidated server.

For example, the business requests to deploy a new application on a new server.
The problem is that you have no server available and need to order one, which
takes time to get, because you need approvals prior to place your order.

In a virtualized consolidated environment, it is as simple as a click to create the

requested server. This happens without any downtime because the hardware is
already there and managed by the system administrators.

Virtualization can be categorized as follows: (see Figure 1-3 on page 10)

› Full hardware and firmware embedded virtualization
With Power Systems, all virtualization functions are implemented into the
hardware and firmware. The firmware is called the Power Hypervisor. See
1.5, “Introduction to PowerVM virtualization” on page 21 for more information
about Power Hypervisor.
› Full virtualization
Full virtualization is a virtualization technique used to provide a certain kind
of virtual machine environment, namely, one that is a complete simulation of
the underlying hardware. In such an environment, any software capable of
execution on the raw hardware can be run in the virtual machine and, in
particular, any operating systems
Full virtualization provides necessary abstractions for the application and
therefore does not require any changes to the application or OS.
› Para virtualization
Para virtualization is a technique that uses an OS-based software interface to
virtual machines and requires OS changes, but offers a performance benefit
on older hardware.
› OS-based virtualization
OS-based virtualization is a technique that allows virtual images to be created
off a single OS instance.
Chapter 1. Introduction 9
 Figure 1-3 Other virtualization approaches

Note: Virtualization concerns the server, the storage, and the network.

1.3 Introduction to AIX 6.1

AIX is an open standards-based UNIX OS that is designed to comply with the
Open Group’s Single UNIX Specification Version 3. It provides the
enterprise-class IT infrastructure for thousands of clients around the world.

On System p servers, mixed environments are supported: AIX 5L V5.2 ML2 and
AIX 5L V5.3 partitions with dedicated processors and adapters, and AIX 5L V5.3
partitions using micro-partitioning and virtual devices. AIX 5L V5.3 partitions can
use physical and virtual resources at the same time.

AIX 5L is supported on the System p servers in partitions with dedicated

processors (LPARs), and shared-processor partitions (micro-partitions).

For System p servers configured with one of the PowerVM features, AIX 5L
Version 5.3 or later is required for micro-partitions, virtual I/O, and virtual LAN.
10 Best Practices for DB2 on AIX 6.1 for POWER Systems
1.3.1 IBM AIX V6.1
AIX V6.1 is the latest version of the AIX operating system, which includes new
and improved capabilities for virtualization, security features, continuous
availability features, and manageability. AIX V6.1 is the first generally available
version of AIX V6.

The following list details on features which AIX V6.1 supports:

› PowerVM AIX 6 Workload Partitions (WPAR), software-based virtualization
› Live Partition Mobility (LPM), with the IBM PowerVM AIX 6 Workload
Partitions Manager™ for AIX
› 64-bit kernel for higher scalability and performance
› Dynamic logical partitioning and micro-partitioning support
› Support for Multiple Shared-Processor Pools
› Trusted AIX, multilevel, compartmentalized security
› Integrated Role Based Access Control
› Encrypting JFS2 (Enhanced Journaled File System 2) file system
› Kernel exploitation of POWER6 Storage Keys for greater reliability
› Robust journaled file system and Logical Volume Manager (LVM) software
including integrated file system snapshot
› Tools for managing the systems environment
› System Management Interface Tool (SMIT) and the IBM Systems Director
Console for AIX

AIX 6.1 runs on systems based on POWER4, PPC970, and POWER5

processor-based plat-forms, but the most capability is delivered on systems built
with the new POWER6 processors.

The AIX OS is designed for the IBM Power System p, System i, System p5®,
System i5®, eServer p5, eServer pSeries®, and eServer i5 server product lines,
as well as IBM BladeCenter® blades based on Power Architecture technology
and IBM IntelliStationPOWER workstations.
Chapter 1. Introduction 11
Table 1-1 provides the AIX V6.1 new features summary information.

Table 1-1 AIX V6.1 new features summary

Features Functionality Benefits

Virtualization

PowerVM Workload WPARs enable the Reduced administration

Partitions creation of multiple virtual and improved system
AIX6.1 environments efficiency.
inside of a singleAIX6.1
instance.

Live Application Mobility Workload Partitions can be Increased application

moved from one system to availability, enhanced
another without restarting workload manageability
the application or causing and energy savings.
significant disruption to the
application user.

Security

Role Based Access RBAC provides improved Improved security,

Control (RBAC) security and manageability decreased administration
by allowing administrators costs.
to grant authorization for
management of specific
AIX6.1resources to users
other than root.

Encrypted FileSystem The IBM Enhanced Improved security.

Journaled Filesystem
Extended (JFS2) adds
even greater data security
with the capability to
encrypt the data in a file
system.

Trusted AIX Trusted AIX extends the Highest level of security for
security capabilities of the critical government and
AIX OS by integrating business workloads.
compartmentalized,
multilevel security (MLS)
into the base operating
system to meet critical
government and private
industry security
requirements.
12 Best Practices for DB2 on AIX 6.1 for POWER Systems
Features Functionality Benefits

Secure by default Enables only the minimal Improved security on initial

(Installation option) number of system and installations of AIX6.1.
network services to
provide the maximum
amount of security.

Trusted Execution Verifies the integrity Improved security.

programs at execution
time.

Support for Long Pass AIX 6.1 and AIX 5.3 Improved security.
Phrases Technology Level 7
supports greater than
eight character passwords
for authentication of users.

Continuous Availability

Concurrent AIX kernel Provides a new capability Improved AIX availability.

update to deliver kernel updates
as interim fixes that do not
require system reboot to
put into effect.

Storage keys Storage keys can reduce Improved AIX availability

the number of intermittent and improved application
outages associated with availability.
undetected memory
over-lays inside the AIX
kernel and kernel
extensions.

Dynamic tracing New tracing command, Easier resolution to

probevue, allows a application execution and
developer or system performance problems.
administrator to
dynamically place probes
in existing application or
kernel code, without
requiring special source
code or even
recompilation.
Chapter 1. Introduction 13
 Features Functionality Benefits

Manageability

IBM Systems Director This new management Reduced administrative

Console for AIX interface allows costs and improved
administrators to manage administrative
AIX6.1 remotely through a effectiveness by enabling
browser. Web-based administration
across multiple AIX
instances.

Automatic Variable Page Automatically manage the Improved performance

Size size of pages used when it with reduced
is running on a system administrative effort.
based on POWER6
processors.

Solution Performance The default tuning Improved performance

Tuning parameters for AIX6.1 with reduced
have been changed to administrative effort.
provide much better
performance for most
applications right out of the
box.

Network Installation Network Installation Improved performance

Manager Support f or Manager (NIM) has been with reduced
NFSv4 enhanced to provide administrative effort.
additional security features
and flexibility by enabling
the use of NFS version4.

In summary, AIX6.1 and POWER6 provides innovative features for virtualization,

security, systems management, and RAS features with no compromise on the
performance.
14 Best Practices for DB2 on AIX 6.1 for POWER Systems
1.4 Introduction to DB2
DB2 Version 9 data server is the fastest-growing, flagship database offering of
IBM. It is optimized to deliver industry-leading performance across multiple
workloads, while lowering administration, storage, development, and server
costs. This section discusses the benefits that DB2 can provide with respect to
these points.

1.4.1 Autonomics
Although costs for server management and administration can be hard to
measure and might be less apparent than costs for servers, storage, and power,
they represent the largest percentage of total IT spending. Refer to Figure 1-4 to
see a breakdown of the various costs associated with administration.

Figure 1-4 Administration costs

Chapter 1. Introduction 15
DB2 can help cut these administrative costs by taking care of itself as much as
possible using the following capabilities:
› Self-configuring:
DB2 automatically sets up the system and manages configuration settings.
For example, using the Configuration Advisor (CA), several instance and
database parameters, including buffer pools, are adjusted to make the
database run well in your environment from the beginning. Using automatic
maintenance, DB2 enables the automation of RUNSTATS, REORG and
BACKUP utilities.
› Self-healing
DB2 automatically helps resolve problems as they occur.
Using the DB2 Health Center it is easy to set up thresholds for warnings, and
alarms for database and instance level parameters. It can recommend the
course of actions to follow to resolve problems. It is possible to prevent
problems and ensure the databases remain always available. Note that the
DB2 9 Health Monitor and Health Center are deprecated in DB2 9.7. The
tools that replace these are available with the IBM Optim™ solutions.
› Self-optimizing
DB2 reacts to changes in workloads and adjusts memory and other facets of
the software to continuously improve performance
The self-tuning memory feature in DB2 9 simplifies the task of memory
configuration by automatically setting values for several memory configuration
parameters at startup. The self-tuning memory manager uses intelligent
control and feedback mechanisms to keep track of changes in workload
characteristics, memory consumption, and demand for the various shared
resources in the database. It dynamically adapts their memory usage as
needed.
› Self-protecting
DB2 addresses external security threats to the system by detecting and
preventing unauthorized access
Extensive security and audit capabilities help protect information from
ever-changing threats, and address regulatory requirements such as the
Sarbanes-Oxley Act. DB2 supports a robust role-based security model,
enabling businesses to divide authority in compliance with data governance
standards.
16 Best Practices for DB2 on AIX 6.1 for POWER Systems
 A combination of the following mechanisms can be used by which DB2 can
provide such extensive security:
– Authentication
– Authorization
– Trusted Contexts
– Auditing
– Label Based Access Control (LBAC)
– DB2 encryption.Compression

DB2 incorporates Deep Compression technology and the DB2 Storage

Optimization Feature, which can reduce storage requirements. For example,
using DB2 row compression, for example, can save up to 83% of disk space on
your largest tables. The most recent release, DB2 9.7, further improves storage
savings by adding data compression for database indexes, temporary database
tables, temporary tables, large objects, and XML documents. Accessing data
from the disk is the slowest database operation. By storing compressed data on
disk, fewer I/O operations need to be performed to retrieve or store the same
amount of data. Therefore, for disk I/O-bound workloads, the query processing
time can be improved noticeably.

Note: The DB2 Storage Optimization Feature is a licensed feature that

enables you to use the DB2 compression features (which already includes
deep compression). Deep compression is often referred to row-level
compression.

1.4.2 pureXML
DB2 supports both relational and XML data, which can simplify development and
deployment of advanced new applications. DB2 pureXML eliminates much of the
work typically involved in the management of XML data, and serves XML data at
unmatched speeds. Applications can mix relational and XML data as business
needs dictate. DB2 9.7 adds end-to-end native XML support for both
transactional and data warehouse applications, opening new opportunities to
extract business value from XML data.

1.4.3 Automatic storage

DB2 9 extends the automated storage features first introduced in DB2 V8.2.2.
Automatic storage automatically increases the size of your database across
disks and file systems. With automatic storage management, DB2 allocates
storage on demand as the table consumption grows. This feature intends to be a
single point of storage management for table spaces. DBAs are no longer
Chapter 1. Introduction 17
 required to define the containers for table spaces, but specify a group of storage
devices for DB2, such as file systems, in the form of storage paths. DB2 creates
the necessary containers automatically across these storage paths. The growth
of the existing containers and the additional new ones is managed by DB2. In
DB2 9, automatic storage is enabled by default for new databases. In addition,
automatic storage support has been added for multi-partition databases.

1.4.4 Performance
The DB2 Performance Optimization Feature gives insight and the ability to
optimize workload execution, which can be accomplished using a combination of
DB2 Workload Manager, DB2 Performance Expert, and DB2 Query Patroller.
This can help reduce hardware acquisition costs by optimizing the performance
of servers and postponing costly hardware upgrades. Figure 1-5 shows that
between January 1, 2003 and March 23, 2009, DB2 has held certain industry
benchmarks for more days than all other vendors combined.

Figure 1-5 DB2 industry benchmarks

18 Best Practices for DB2 on AIX 6.1 for POWER Systems
1.4.5 Reliability and scalability
DB2 helps businesses increase availability by reducing planned and unplanned
outages. There are a few methodologies that can be used to ensure
uninterrupted access to the database, namely DB2 Replication (also known as
DataPropagator), High Availability Disaster Recovery (HADR) and DB2
pureScale. DB2 HADR allows failover to a standby system in the event of a
software or hardware failure on the primary system. Online reorganization
actively reconstructs the database table for better performance, while permitting
uninterrupted access to the table data. DB2 pureScale, introduced in DB2 9.7
Enterprise Server Edition, offers clustering technology that helps deliver high
availability and exceptional scalability transparent to applications.

Unlike other distributed shared-disk database cluster technologies, DB2

pureScale does not require administrators to perform complex tuning or update
application code when scaling. Flexible application workload balancing and
grouping capabilities allow multiple servers to appear as a single database. New
members (DB2 server instances) can be used immediately. It is designed to
simplify database and clustering administration so that members can be added
to and removed from the cluster easily, enabling IT staff to scale the cluster up or
down quickly to meet changing business requirements. DB2 pureScale also
delivers enhanced performance by using IBM Power Systems and incorporating
PowerHA pureScale technology. Figure 1-6 shows the DB2 pureScale.

Figure 1-6 DB2 pureScale

Chapter 1. Introduction 19
1.4.6 Multi-vendor SQL and API support and migration to DB2
There are number of enhancements that simplify the movement of applications to
DB2 and use existing skills. With DB2 9.7, differences are now the exception, not
the rule, when it comes to many database and development features, including
concurrency models, SQL dialects, data types, procedural languages, packages
and scripting languages. Applications moved to DB2 can run with full native
execution, delivering high performance.

For more information see the SQL compatibility enhancements document at the
following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/topic/com.ibm.db2.
luw.wn.doc/doc/c0054107.html

1.4.7 DB2 in virtualization environment

Coupled with the IBM PowerVM features, DB2 9.7 is uniquely positioned to
exploit virtualization. DB2 9 is enabled to use advanced virtualization features
that make it an excellent choice for an information-management platform on the
Power System server. DB2 works seamlessly in the Power system virtualization
environment. There is no additional package or driver to install or special
configuration required for either DB2 or AIX. Additionally, DB2 is aware of, and
reacts to, any dynamic logical partitioning (DLPAR) event (that is, any runtime
change to host-partition operating-system resources, such as a change in
computing resources and physical memory). The self-tuning
memory-management (STMM) feature of DB2 9 automatically adjusts and
redistributes DB2 heap memory in response to dynamically changing partition
memory and workload conditions. The STMM feature, coupled with LPM, allows
you to move an active database that runs on a system with limited memory
resources to a system that has additional memory. Then, with the help of DLPAR
operation and STMM, the database can make use of this additional memory,
providing increased throughput to the workload.

Given this, IBM has introduced sub-capacity (virtualization) licensing. This

enables customers to use server virtualization to consolidate their infrastructure
more effectively and reduce their overall total cost of ownership (TCO). It also
allows flexible software licensing using advanced virtualization capabilities (such
as shared processor pools, micro-partitioning, virtual machines and dynamic
reallocation of resources). This new licensing gives expanding customers the
flexibility to choose how to add workload environments without making trade-offs
between hardware design and software licensing.
20 Best Practices for DB2 on AIX 6.1 for POWER Systems
1.5 Introduction to PowerVM virtualization
PowerVM is part of the IBM Advanced Power Virtualization offering and is
available on Power Systems-based servers. PowerVM helps confine your
investment to your hardware. It is a combination of elements (such as hardware,
firmware and software) that provide virtualization for your CPU, your network,
and your disks.

Partitions can have dedicated or shared processor resources. With shared

resources, PowerVM can automatically adjust pooled processor resources
across multiple operating systems, borrowing processing power from idle
partitions to handle high transaction volumes in other partitions.

Key components are as follows:

› The hardware
POWER6 hardware systems provide the hardware component of PowerVM or
POWER6 hardware systems are the hardware component of PowerVM.
› Power Hypervisor
Power Hypervisor is the foundation for virtualization on Power Systems,
enabling the hardware to be divided in multiple logical partitions, and ensuring
complete isolation between them. Power Hypervisor is designed to provide
business-critical availability. It is always active and cannot be de-activated. It
is responsible to dispatch the workload across the physical shared
processors, and provide inter-partition communication enabling the VIOS’s
virtual SCSI and virtual Ethernet function.
› VIOS
VIOSs are enhanced AIX servers with the capability to virtualize I/O adapters,
such as Ethernet cards and Fibre Channel cards. The virtualization of those
elements enables you to benefit fully from your hardware investment.

Moreover, PowerVM extends the base system functions of your server to include
the following capabilities:
› Micro-partitioning
Micro-partitioning enables a processor to be divided to a tenth of a whole
CPU and is available from POWER5 systems, running AIX 5.3 or later. It adds
flexibility to manage your processors efficiently to create virtually partitions up
to ten times the number of available processors.
Chapter 1. Introduction 21
› Live partition mobility
This capability enables you to move a running logical partition (known as
Active partition mobility) physically from one physical server to another,
without interrupting application service (interruption of some milli-seconds
only) to your clients. Inactive partition mobility is also possible for those
partitions that are in power off mode. Live partition mobility improves the
availability of your server by eliminating planned outages, and balancing
workloads during peaks.
› (Dynamic) logical partitioning (DLPAR)
A logical partition is a set of system resources that contain whole or a portion
of a CPU, memory, physical, or virtual I/O resources, all logically grouped into
one same partition. The dynamic capability allows adding resources
dynamically and on demand.
› Virtual Ethernet
Virtual Ethernet is managed by the Power Hypervisor and shares the internal
Power Hypervisor bandwidth.
› Shared Ethernet Adapter (SEA)
SEAs are logical adapters that bridge a physical Ethernet card with the virtual
Ethernet adapter, so that a logical partition can communicate with the outside
world.
› Shared-processor pools
Shared processor logical partitions are allocated processor units out of the
processor pool. The amount of processor capacity that is allocated to that
partition (its entitled capacity) can be as small as a tenth of a physical CPU
and as big as the entire processor. The granularity to add processor units is
1/100 of a CPU. The shared processor pool can have up to one whole CPU
contained in the physical server.
› N-Port Identifier Virtualization (N-PIV)
N-PIV adapters are particular 8 GB Fibre Channel adapters that allow a client
to see the SAN device, transparently provided by the VIOS. For example the
client logical partition sees the hdisk as MPIO DS5100/5300 Disk, instead of
as a Virtual SCSI Disk Drive.
22 Best Practices for DB2 on AIX 6.1 for POWER Systems
PowerVM provides advanced virtualization capabilities to your Power System or
Power Blade server. It exists in editions as shown in Figure 1-7.

Figure 1-7 PowerVM Editions

› Express Edition
The Express ASE edition allows you to use only 1 VIOS and a single LPAR,
managed by the Integrated Virtualization Manager (IVM).
› Standard Edition
This edition allows you to create as much as 10 LPARs per core, and
supports multiple shared processors pools. See the following Web page for
limits:
http://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/TD103130
› Enterprise Edition
This edition covers all capabilities of the express and standard edition, plus
the Live Partition Mobility and the Active Memory™ Sharing capabilities.
Chapter 1. Introduction 23
All Power Systems servers can use standard virtualization functions or logical
partitioning (LPAR) technology by using either the Hardware Management
Console (HMC) or the Integrated Virtualization Manager (IVM). Figure 1-8 shows
the PowerVM virtualization architecture.

Figure 1-8 PowerVM virtualization architecture

The HMC is a separate server that controls the build of logical partition (LPAR),
enables starting and stopping of logical partitions at distance by a remote
terminal console, and enables the dynamic functionalities of your Power System
to add, move or remove hardware to your logical partitions (DLPAR) dynamically.
An HMC can manage multiple Power Systems. The HMC user interface is
Web-based and is the focal point for your server environment and the hardware.

Note: The HMC is only used with Power Systems servers, while the Integrated
Virtual Manager (IVM) is only used for Power Blades servers.

Similar to the HMC, the IVM can be used to create and maintain logical
partitions. It is also integrated with the VIOSs. Nevertheless, the IVM can only
control one Power System.

Note: IVM is disabled if an HMC is connected to your Power System.

24 Best Practices for DB2 on AIX 6.1 for POWER Systems
 LPAR: logical partition are built out of a profile. It can be seen as a virtual
machine. The entire Power System can be divided into one single or multiple
LPARs, each running its own operating system and completely isolated from
each other. The profile contains all of the information needed to boot the logical
partition, number of processors, memory, and I/O adapters (physical or virtual).

Note: A logical partition’s profile can be configured to contain all of a system’s

resources. This is similar to a standalone system.

Tip: Rather than configuring your system as a full system partition, you may
consider configuring a partition containing all the system resources. This gives
you more flexibility if you want to add an extra partition to your system.

1.5.1 Logical partition type

Logical partitions are built upon a profile that contains all the physical and virtual
devices that you need to start your partition. In Figure 1-9 on page 26, you see
the flow to create a profile for a logical partition. Frequently, questions arise
regarding whether a dedicated or shared processor is the correct logical partition
type for a particular workload and system environment. There are a number of
factors to consider, such as performance, overall system use, and consolidation
of multiple workloads onto a single server to improve power usage and cooling
efficiencies. The chart in Figure 1-9 on page 26 shows a macro view of how you
can configure a logical partition’s profile.

For more information refer to Chapter 5, “LPAR considerations” on page 225.

Chapter 1. Introduction 25
Figure 1-9 Profile chart

1.5.2 Processor virtualization

When dividing a Power System, you can choose between two partitioning
methods depending on how you want to share the CPU capacity among the
LPARs:
› Shared
› Dedicated

Figure 1-10 on page 27 shows the terminologies used when talking about the
processor distribution for your LPARs.
26 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 1-10 Processors terminologies

Let us go over a few of these virtualization concepts:

› Dedicated CPU
A dedicated CPU is an entire processor that you allocate to your profile. Only
that particular LPAR uses that processor.

Attention: Upon activation, if a logical partition does not use 100% of its
dedicated processor resources, the unused processor resources are ceded to
the shared processor pool. This is called Donating Dedicated CPU. When the
logical partition needs that donated CPU resource back, the Power Hypervisor
immediately liberates it and give it back to the donating logical partition.

› Shared CPU
With shared CPUs, physical CPUs are placed in a pool of processors and
shared between multiple logical partitions. The logical partition is assigned a
portion of that CPU pool. It can be as small as a 10th of a real CPU, with a
granularity of a 100th. That portion is called a processing unit. The ability to
divide a CPU into smaller units is known as the Micro-Partitioning
technology. The granularity of that processor unit is a hundredth of a CPU. An
example of a shared processor unit allocation is 1.34 CPU. In this mode,
processing units that are not used return to the shared processor pool for
other LPAR if they need additional CPU.
Chapter 1. Introduction 27
 Tip: In shared CPU mode, the logical partition is guaranteed to have its
processing entitlement whenever it needs it. The Power Hypervisor is
managing the processing entitlements and allocates it to the demanding
partition.

With the shared CPU, we have two other configurations:

– Capped mode
In capped mode, the processor units allocated to the logical partition
cannot exceed the processor unit it was allocated.
– Uncapped mode
In uncapped mode, the logical partition can get more CPU units than it is
allocated, up to the shared CPU pool available units. These additional
CPU units are given and treated by the Power Hypervisor. In this mode, a
weight factor prioritizes the portion of CPU that is given to a partition, as
shown in Figure 1-11 on page 29.
In Figure 1-11 on page 29, the following abbreviations are used:
• EC stands for Entitled Capacity and represents the actual processor
core capacity available to a partition.
• VP stands for Virtual Processor and defines the maximum number of
physical processor cores that the system can access simultaneously to
provide the processing capacity of a shared processors partition.
28 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 1-11 Weight factor calculation

› Active processors
Active processors are the usable physical processors in the system.
› Inactive processors
Inactive processors are those processors that can be activated on demand.
They are known as Capacity On Demand (COD) processors and need an
activation code to be converted into active processors
› Deactivated processors
Deactivated processors are processors that have a hardware problem. In
case of a hardware problem and if COD processors are available, the broken
processor is swapped with a COD processor. If COD processors are not
available, ask an IBM Customer Engineer engineer to change the defective
processor, which can require down-time.
› Virtual processors
Virtual processors are abstractions of physical processors that are assigned
to logical partitions. The operating system uses the number of virtual
processors assigned to the logical partition to calculate the number of
Chapter 1. Introduction 29
 operations that the operating system can perform concurrently. Virtual
processors are relevant to micro-partition. Micro-partitioning it maps virtual
processors to physical processors. The virtual processors are assigned to the
partitions instead of the physical processors.
› SMT Simultaneous Multi-Threading
SMT Simultaneous Multi-Threading is a concept where multiple threads of
execution can execute on the same processor at the same time. With SMT
enabled, each hardware thread is seen as a logical processor. SMT is not a
virtualization concept.

Tip: When SMT is not enabled, a virtual processor appears as a single logical
processor.

1.5.3 Memory virtualization

PowerVM has been enhanced with active memory sharing (AMS), an advanced
virtualization technology that intelligently flows memory from one logical partition
to another to increase use and flexibility of memory. It allows for dynamic sharing
of memory. It is designed for logical partitions with variable memory needs. Using
AMS improves the overall memory use of your Power System, as shown in
Figure 1-12 on page 31. The graph shows the percentage of memory assigned
to a partition at a particular point of time for dedicated memory for AMS.
30 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 1-12 Dedicated versus shared memory environment

Note: AMS needs PowerVM Enterprise Edition key installed on your system.

› Dedicated memory
Dedicated memory is physical memory dedicated to an LPAR. It is reserved
for that partition and cannot be shared. A logical partition that uses dedicated
memory, known as a dedicated memory partition, only uses a fixed amount of
memory that it was assigned.
› Shared memory
Shared memory is that memory assigned to the shared memory pool and
shared among multiple logical partitions. A logical partition that uses shared
memory, known as a shared memory partition, allows an appropriate logical
memory size to be defined without requiring a corresponding amount of
physical memory to be allocated.

You must specify micro-partitions if you want to specify shared memory for this
logical partition.
Chapter 1. Introduction 31
1.5.4 I/O virtualization
Virtual I/O is often referred to as a set of network and storage virtualization
features, such as:
› Virtual Ethernet
The Virtual Ethernet adapter allows logical partitions to communicate with
each other within the same Power System, using the Power Hypervisor’s
internal switch. They do not need any network hardware adapter or cables to
communicate.
› Shared Ethernet Adapter (SEA)
This logical adapter enables an LPAR to communicate with the outside world.
SEA adapters can be configured in a SEA fail-over mechanism to protect your
client logical partitions from VIOS network failure. Moreover, SEA is
mandatory for Live Partition Mobility use.
› Integrated Virtual Ethernet (IVE)/Host Ethernet Adapter (HEA)
IVE, or HEA, is a dual or quad port Fibre Channel card, available in 1 Gbps
(dual or quad port) or 10 Gbps (quad port only). It enables an easy way to
manage the sharing of the integrated high-speed Ethernet adapter ports. The
integrated virtual Ethernet is directly connected to the GX+ bus instead of
being connected to a PCIe or PCI-X bus. This provides the IVE card with high
throughput and low latency. It is available on p520 and p550 servers only. The
IVE card is installed by manufacturing and does not support hot-swappable or
hot-plugable capabilities.

Figure 1-13 illustrates the difference between the use of SEA and IVE.

Figure 1-13 SEA versus IVE use

32 Best Practices for DB2 on AIX 6.1 for POWER Systems
› Virtual storage
The virtual SCSI adapter allows logical partitions to connect to the SAN
storage through the VIOSs. It is recommended to configure your virtual server
SCSI adapters to connect to the virtual client SCSI adapter. In this way you
ensure that only that particular logical partition can connect to the disks
mapped on the VIOS. LUNs from VIOSs can be mapped in entire disks or
masked as LUN (a logical volume is first created on the VIOS and then
masked to the client).
› N_Port ID Virtualization
N_PIV adapters simplify the management of Fibre Channel SAN
environments. They enable access to SAN devices. NPIV allows multiple
N_Port IDs to share one single physical N_Port. This allows multiple Fibre
Channel initiators to occupy a single physical port. N-PIV allows the SAN
based LUN to be exported to the LPAR rather than the VIOS. This puts the
SAN administrators back in charge of which LPAR owns the LUNs, so they
can decide which LPAR gets low cost disks or high performance disks.
Currently, they give the LUN to the VIOS and the VIOS administrator decides
which LPAR gets the disk space. N-PIV adapters are physical 8 GB Fiber
Cards and require N-PIV support on your switches.

All virtual network and storage devices are identified with their slot number. Slot
number can aid you in your management while creating your naming convention,
for situations where lots of logical partitions co-exists and where administration of
those virtual resources become more and more complex.

Tip: In a dual VIOS setup, try to get a mapping in slot numbers to ease the
management of your servers. For example, you can use a particular slot
number range dedicated to your virtual networks and make it big enough to
allow for growth (for example 11–30) where virtual SCSI slot number range
ranges from 31 to 50.
Chapter 1. Introduction 33
1.5.5 Useful links
The following list the links to the Information Center documentation for the main
topics discussed in this chapter. You can refer to these links for more information
about these topics.

AIX 6.1
› http://publib.boulder.ibm.com/infocenter/aix/v6r1/index.jsp

› http://publib16.boulder.ibm.com/pseries/index.htm

› http://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/TD103130

PowerVM
› http://www-03.ibm.com/systems/power/software/virtualization/index.html
› http://www14.software.ibm.com/webapp/set2/sas/f/vios/documentation/ins
tallreadme.html

Power Systems
› http://publib.boulder.ibm.com/infocenter/powersys/v3r1m5/index.jsp

› http://www-03.ibm.com/systems/power/hardware/

› http://www14.software.ibm.com/webapp/set2/sas/f/power5cm/power6.html

DB2
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/topic/com.ibm.db2
.luw.wn.doc/doc/c0054107.html
34 Best Practices for DB2 on AIX 6.1 for POWER Systems
 2

Chapter 2. AIX configuration

This chapter discusses the recommended AIX tunable for the best performance
of DB2. By default, AIX V6.1 provides the optimized tunable values for the best
performance of the system and application. DB2 recommends certain
configurable parameter changes as the best practise taking the performance and
security into considerations. The DB2 registry variables on AIX is also discussed.
Finally, a table containing the configurational parameters differences between
AIX 5.3 and AIX 6.1 release is provided for quick reference.

Topics discussed in this chapter are as follows:

› “AIX configuration and tuning for DB2” on page 36
› “DB2 registry variables” on page 72
› “Configurational differences: AIX 5.3 versus AIX 6.1” on page 80
© Copyright IBM Corp. 2010. All rights reserved. 35
2.1 AIX configuration and tuning for DB2
In this section we discuss AIX tunable parameters germane to DB2.

2.1.1 Virtual memory considerations

AIX Virtual Memory Manager (VMM) classifies memory into computational and
non-computational memory.
› Computational memory includes application memory (for example, DB2
process text, data, and stack segments) and kernel memory.
› Non-computational memory is classified as file system cache, which includes
file system data from JFS (journaled file system), JFS2, NFS (Network File
System), GPFS™ (General Parallel File System) or any file system type. This
is also considered as permanent memory.
– All permanent memory except JFS is called client memory.
– All file system cache is tracked by the parameter, numperm.
– Size of the client memory is tracked by the VMM parameter, numclient.

Figure 2-1 shows the computational and non computational memory occupancy
in real memory.

Working
Computational Paging
Memory Space
Segment

Persistent
Non computational File System,
permanent memory JFS
Segment

Non computational Client Client Pages

Client memory NFS , JFS2
Segment
Real Memory

Figure 2-1 Computational and non-computational memory occupancy

36 Best Practices for DB2 on AIX 6.1 for POWER Systems
Computational memory is represented as working storage (segment) in the
svmon output. Non computational memory is considered to be either client or
persistent segment.This been represented as “clnt” in the svmon output as
shown in Example 2-7 on page 45. This is also called file pages.

Figure 2-2 shows the file system cache consumption in real memory.

maxperm
90 %

Computational
numperm pages

Non computational
minperm File
System cache
3%

Real Memory
Figure 2-2 The file system cache consumption as pointed out by numperm

minperm and maxperm values show the range that the file system cache can
grow. If computational pages need more memory, then the file cache pages are
stolen (page stealing) based on the repaging rate.

When there is no record of the page having been referenced recently, it is

considered as a new page fault. But when a page that is known to have been
referenced recently is referenced again and not found in memory, it is considered
as repage fault. Keeping in control of the repage faults reduces I/O demand and
potentially improve the system performance.

lru_file_repage specifies whether the repaging rate is considered in determining

whether to steal a file or computational pages, when the file pages consumption
goes beyond minperm.
Chapter 2. AIX configuration 37
 VMM maintains a logical free list of memory pages to satisfy a page fault. It uses
a page replacement algorithm to discover which memory pages (computational
or non computational) currently in memory are to be moved to free list.
› minfree specifies the number of memory frames in the free list at which VMM
starts to steal pages in memory to replenish the free list.
› maxfree specifies the number of memory pages in the free list at which VMM
stops page stealing.
› Page replacement is done through the kernel process “lrud”.

Figure 2-3 shows the page stealing based on minfree and maxfree values.

Figure 2-3 Page stealing based on minfree and maxfree values

2.1.2 AIX vmo command

This command sets or displays the current or next boot values for all VMM tuning
parameters. This command can also make changes transient (lost at the next
boot) or permanent (applied at the next boot).

For example, the vmo -L command displays the current, default, and the boot
vmo parameter settings, as shown in Figure 2-4 on page 39.
38 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 2-4 Output of vmo -L

Example 2-1 shows the current vmo settings in the system.

Example 2-1 Output of “vmo -a”

# vmo -a
cpu_scale_memp = 8
data_stagger_interval = 161
defps = 1
force_relalias_lite = 0
framesets = 2
htabscale = n/a
kernel_heap_psize = 4096
kernel_psize = 4096
large_page_heap_size = 0
lgpg_regions = 0
lgpg_size = 0
low_ps_handling = 1
lru_file_repage = 0
lru_poll_interval = 10
lrubucket = 131072
Chapter 2. AIX configuration 39
 maxclient% = 90
maxfree = 5418
maxperm = 219580
maxperm% = 90
maxpin = 212018
maxpin% = 80
...........

Example 2-2 shows how to modify and make the changes persistent across the
reboot. The vmo -o command is used to make the changes, and the option -p is
used to make the changes persistent dynamically and across reboot.

Example 2-2 vmo to make the changes permanent across reboot

# vmo -p -o minfree=4906 -o maxfree=5418

Setting maxfree to 5418 in nextboot file Setting
minfree to 4906 in nextboot file Setting
maxfree to 5418
Setting minfree to 4906

Note: Beginning with AIX Version 6.1, tunables are classified as restricted use
tunables. They exist and must be modified primarily for specialized
intervention by the IBM development support or development teams.

As these parameters are not recommended for user modification, they are no
longer displayed by default but only with the -F option (force) on the command.
For example, the command vmo -F -a lists all the tunables including
restricted.

2.1.3 VMM considerations for DB2

Protecting computational memory is important for DB2, as it maintains its own
data cache. Following AIX tunables discusses about the default values and the
DB2 recommended values.

minfree
When the size of the free list falls below this number, the VMM begins stealing
pages. It continues stealing pages until the size of the free list reaches maxfree.
40 Best Practices for DB2 on AIX 6.1 for POWER Systems
If the value free frame waits from vmstat -s value increases over the period of
time, then it is advisable to increase the minfree value.
› Default value in AIX6.1: 960
› DB2 recommended value:
– 4096 for memory less than8 GB
– 8192 for memory greater than 8 GB
› How to set: vmo -p -o minfree=4096
› Type: Dynamic (No reboot required)

maxfree
The difference between minfree and maxfree must not go beyond 1024, as it
increases the time in replenishing the free list. The difference must be equal to or
greater than j2_maxPageReadAhead common maxpgahead IO tunable value.
See 2.1.7, “Input and output tunable considerations” on page 63 for details about
this tunable. See Example 2-3.
› Default Value in AIX6.1: 1088
› DB2 Recommended value:
– minfree + 512
– if minfree is 4096, then maxfree=4608
› How to set: vmo -p -o maxfree=4608
› Type: Dynamic

Example 2-3 Setting the maxfree value

# vmo -p -o maxfree=4608 -o minfree=4096

Setting maxfree to 4608 in nextboot file
Setting minfree to 4096 in nextboot file Setting
maxfree to 4608
Setting minfree to 4096

maxclient%
This specifies the maximum physical memory usage for client pages caching. If
the percentage of real memory occupied by file pages is higher than this level,
the page-replacement algorithm steals only client pages. It is recommended to
set strict_maxclient to 1 along with this setting for DB2. This considers maxclient
as the hard limit and strictly steals client pages as soon as the limit is reached.
› Default value in AIX6.1: 90%
› DB2 recommended value: 90%
Note: This is a restricted tunable in AIX6.1. Hence no further change is
recommended.

Chapter 2. AIX configuration 41

maxperm%
This specifies the maximum physical memory usage for file pages caching.

If the percentage of real memory occupied by file pages rises above this level,
the page-replacement algorithm steals only file pages.
› Default value in AIX6.1: 90%
› DB2 recommended value: 90%

Note: This is a restricted tunable in AIX6.1. Hence no further change is

recommended.

minperm%
If the percentage of real memory occupied by file pages falls below this level, the
page replacement unit steals only computational values. Hence a low value is
recommended for DB2. AIX by default sets this value to 3% compared to AIX5L
where by default it is 20%.
– Default value in AIX6.1: 3%
– DB2 recommended value: 3%

Note: This is not a restricted tunable in AIX6.1.

strict_maxclient
If the client page caching increases beyond the maxclient% value, setting this
parameter acts as a hard limit and triggers file pages to be stolen.
– Default value in AIX6.1: 1
– DB2 recommended value: default value

Note: This is a restricted tunable in AIX6.1. Hence no further change is

recommended.

lru_file_repage
When VMM needs memory if the free memory pages goes below minfree, or
when maxclient% is reached with strict_maxclient set, the lru makes a decision
whether to steal the computational pages or the file pages.

This determination is based on the number of parameters, but the key parameter
is lru_file_repage.
42 Best Practices for DB2 on AIX 6.1 for POWER Systems
Setting this tunable value to 0 indicates that only the file pages are stolen if the
file pages are greater than the minperm value. This prevents computational
pages from being paged out. This is a restricted tunable in AIX6.1. Hence no
further change is recommended.
› Default value in AIX6.1: 0
› DB2 recommended value: default value

Notes:
› AIX Version 6 allows the system to use up to 90% of its real memory for file
caching, but it favors computational pages as resident pages over file
pages.By setting lru_file_repage to 0, it forces the page replacement
algorithm to steal computational pages only when the percentage of
cached file pages is less than the minperm% value. Hence it might not be
required to reduce the maxclient, maxperm value to avoid paging as
practiced with AIX5.3.
› In the situation where reducing the maxperm and maxclient values to lower
helps to improve the performance, it is recommended to inform or contact
IBM Technical Support for a complete solution. IBM does not support any
changes to restricted tunables where the change is the solution.
› Tunables such as lru_poll_interval and strict_maxperm are restricted
tunables and DB2 recommends the AIX6.1 default value. No change is
suggested.
› The tunable page_steal_method is restricted tunable. No change is
suggested. Its default value is 1, which creates a linked list of dirty pages to
steal the pages rather than scanning memory.
Chapter 2. AIX configuration 43
Steps to monitor the memory
The following describes the steps to monitor the memory:
1. The svmon command can be used to monitor the computational memory (such
as the stack heap of a process) and the persistent memory (such as the file
system cache). Computational memory is shown as work, which it considers
as working storage, the persistent memory is shown as, pers, and the paging
space as pgsp. Example 2-4 shows the svmon -G output.

Example 2-4 svmon -G

# svmon -G
size inuse free pin virtual
memory 524288 422096 171824 206465 330676
pg space 131072 3014
work pers clnt other
pin 118951 0 0 87514
in use 330676 0 91420
PageSize PoolSize inuse pgsp pin virtual
s 4 KB - 276528 3014 103585 185108
m 64 KB - 9098 0 6430 9098

In Example 2-4, most of the memory is consumed by computational memory,

330676 4 K pages. Persistent memory is nil and client memory has consumed
91420 4 K pages. Also, only a few 4 KB pages have consumed the paging space.

As the maxperm and minperm values are restricted and not recommended to
change in AIX6.1, it is suggested to monitor the memory usage as described in
the steps that follow:
1. Use lsps -a to understand the paging space consumption over the period of
time. See Example 2-5.

Example 2-5 lsps -a

# lsps -a
Page Space Physical Volume Volume Group Size %Used Active Auto Type Chksum

hdisk0 rootvg 512MB 3 yes yes lv 0

2. Monitor the paging activity using vmstat, as shown in Example 2-6.

Example 2-6 vmstat-Monitor pi/po for paging activity

# vmstat 1

System configuration: lcpu=6 mem=2048MB ent=0.50

 kthr memory page faults cpu

44 Best Practices for DB2 on AIX 6.1 for POWER Systems

----- ----------- ------------------------ -----------------------------------
r b avm fre re pi po fr sr cy in sy cs us sy id wa pc ec
1 0 330629 168041 0 0 0 0 0 0 8 232 190 0 1 99 0 0.01 1.8
1 0 330629 168041 0 0 0 0 0 0 8 93 176 0 1 99 0 0.01 1.3
1 0 330793 167877 0 0 0 0 0 0 10 6769 177 7 7 86 0 0.07 14.5

Note: Use svmon -Pgt 5 to understand the top 5 paging process

3. Use the vmstat command to understand the file cache usage by the system,
as shown in Example 2-7.
Example 2-7 vmstat -v | grep num
vmstat -v | grep num
17.6 numperm percentage
17.6 numclient percentage

4. Run svmon -Put 5 to understand which process is consuming more memory,

which lists the top five processes. See Example 2-8.

Example 2-8 svmon -Put 1

Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd 16MB

266418 java 42548 7663 0 37138 N Y N

PageSize Inuse Pin Pgsp Virtual

s 4 KB 21748 79 0 16338
m 64 KB 1300 474 0 1300

Vsid Esid Type Description PSize Inuse Pin Pgsp Virtual

3449 3 work working storage sm 15677 0 0 15677
3205d d work fork tree m 792 0 0 792
children=4d6cdc, 0
4002 0 work fork tree m 508 474 0 508
children=802760, 0
23199 s 694 0 - -
- clnt /dev/hd9var:12359
ff6f - clnt /dev/hd9var:16629 s 676 0 - -
173a3 - work s 381 75 0 381
3f437 f work working storage sm 247 0 0 247
2b39d - clnt /dev/hd9var:12300 s 243 0 - -
33591 - clnt /dev/hd9var:20583 s 230 0 - -
1d406 - clnt /dev/hd9var:16392 s 212 0 - -
d0ee - clnt /dev/hd9var:20700 s 208 0 - -
192c4 - clnt /dev/hd9var:12363 s 178 0 - -
37573 - clnt /dev/hd9var:28845 s 164 0 - -
17863 - clnt /dev/hd9var:28853 s 134 0 - -
191a4 - clnt /dev/hd9var:12373 s 104 0 - -
2b11d - clnt /dev/hd9var:12379 s 94 0 - -
195e4 - clnt /dev/hd9var:20991 s 83 0 - -
3d7f6 - clnt /dev/hd9var:28797 s 61 0 - -
Chapter 2. AIX configuration 45
 237d9 - clnt /dev/hd9var:28860 s 61 0 - -
1d666 - clnt /dev/hd9var:32821 s 60 0 - -
21398 - clnt /dev/hd9var:12370 s 57 0 - -
3b5d5 - clnt /dev/hd9var:16585 s 53 0 - -
376d3 - clnt /dev/hd9var:28838 s 51 0 - -
39574 - clnt /dev/hd9var:24714 s 50 0 - -
75ab - clnt /dev/hd9var:16580 s 50 0 - -
3f737 - clnt /dev/hd9var:28738 s 50 0 - -
255da - clnt /dev/hd9var:20987 s 49 0 - -
1f667 - clnt /dev/hd9var:32822 s 44 0 - -
3b0f5 - clnt /dev/hd9var:12382 s 40 0 - -

Note:
› If DB2 is doing more file cache, determine if there are any table spaces set
up with file system cache, and decide whether file system caching can be
disabled.
› If there is any non-DB2 file systems or process causing file system cache
to increase and causing paging leading to low performance, rectify it.
› To determine which table spaces have file system cache enabled, use the
db2pd -db <database name> -tablespaces command and see the FSC
column. The possible values are ON or OFF.

Monitoring helsp to understand the memory activity in the system. Because AIX
does not recommend changing the restricted values such as minperm,
maxperm, contact IBM technical support for further assistance, if required. In
summary, the best practise for DB2 regarding AIX VMM setting is as follows:
› minfree=4096 for physical memory <8 GB and 8192 for memory > 8 GB
› maxfree=minfree+512
› maxperm%=90 (default)
› maxclient%=90(default)
› strict_maxclient=1(default)
› minperm%=3(default)
› lru_file_repage=0(default)

Except for minfree and maxfree, no other modification is required. It does not
require reboot of the system after modification.

2.1.4 Large page considerations

Using larger page sizes increases memory access performance because fewer
address translations in the hardware have to be done, and caching mechanisms
can be used more efficiently. However, memory regions might be wasted if a
larger page size is allocated and populated with data less than the page size.
46 Best Practices for DB2 on AIX 6.1 for POWER Systems
Starting with AIX6.1 on a POWER6-based processors the VMM can dynamically
promote pages to a larger page size. This page promotion (4 k to 64 k) is
completely transparent to the application and is done without the need of user
intervention. To see how many large pages are in use on your system, use the
vmstat -l command.

Note: VMM only supports dynamically varying the page size of working
storage memory. Working storage memory comprises process stack, data and
shared library text segment in AIX address space.

To make use of the 16 MB page size, you have to specify the amount of physical
memory that you want to allocate to back large pages. The default is not to have
any memory allocated to the large page physical memory pool. It can be
specified using the vmo command. The following example allocates 1 GB to the
large page physical memory pool (a reboot is required after running the
command to make the changes effective):
vmo -r -o lgpg_regions=64 -o lgpg_size=16777216

Pros and cons of enabling a large page

The following list details the pros and cons of enabling a large page:
› AIX treats large pages as pinned memory. Therefore, no paging support is
available.
› The data of an application that is backed by large pages remains in physical
memory until the application completes.
› Enabling 16 MB pages blocks the memory for its usage. This deprives the
application of using that pool of memory, which affects the system if it does
not have sufficient physical memory for other application requirement.
› Memory allocated to large page is not available for 4 k. Therefore, allocating
too much physical memory to a large page degrades the system.
› The size of the large page physical memory pool is fixed at boot time and
remains the same. A reboot is required to change the size of the large page
memory pool.
› Applications which do a large number of fork()s (such as shell scripts) are
especially prone to performance degradation when large pages are used.

Note: Huge memory pages cannot be used if Live Partition Mobility is

considered (unless you are planning to use inactive mobility).
Chapter 2. AIX configuration 47
Large page enablement for DB2
This is an optional setting for DB2. If your system has sufficient physical memory
such that allocating huge memory for large page does not affect 4 k page
allocation, then enabling 16 MB page can be considered.

To enable the 16 MB page for DB2, follow these steps:

1. Issue the vmo command with the following flags:
vmo -r -o lgpg_size=LargePageSize -o lgpg_regions=LargePages
In the example, LargePageSizespecifies the size in bytes and LargePages
specifies the number of large pages to reserve.
For example, if you need to allocate 25 GB for large page support, run the
command as follows:

vmo -r -o lgpg_size=16777216 -o lgpg_regions=1600

2. Run the bosboot command so that the vmo command takes effect following the
next system boot.
bosboot -ad /dev/ipldevice
shutdown -Fr now
3. After the server comes up, enable it for pinned memory. Issue the vmo
command with the following flags:
vmo -o v_pinshm=1
4. As root user, add the following capabilities for the DB2 instance user to
prevent unauthorized users from using large pages for their applications.
chuser capabilities=CAP_BYPASS_RAC_VMM,CAP_PROPAGATE $USER
5. Use the db2set command to set the DB2_LARGE_PAGE_MEM registry
variable to DB, then start the DB2 database manager. For example:
db2set DB2_LARGE_PAGE_MEM=DB
db2start
6. Validate large page support with the following command:
vmstat -l 1 or vmstat -P ALL

Note: The tprof -a -y <process> -O all command can also be used for
large page analysis.
48 Best Practices for DB2 on AIX 6.1 for POWER Systems
2.1.5 Paging space considerations for DB2
We recommend that DB2 be tuned in a way that does not use paging space
regularly. Periodic paging space usage monitoring is recommended. The
following recommendations are for the paging space considerations:
› The general recommendation is to have the paging space be twice the size of
the physical memory.
Example: For a RAM size of 128 GB it is recommended to have 256 GB
paging space.
› Use a minimum paging space at default rootvg under the size of the RAM not
less than 512 MB and then have the additional (multiple) paging space in
alternate disks not exceeding 64GB.
› Use multiple paging spaces, each allocated from a separate physical volume.
More than one paging space on a disk is not recommended.
› Create the secondary paging space on physical volumes that are more lightly
loaded than the physical volume in rootvg. If minimum inter-disk policy is
chosen, moving the paging space to another disk in the same volume group is
fine.
› The secondary paging spaces must all be of the same size to ensure that the
algorithm performed in turn can work effectively.
› It is better to have the paging space on fast local disks. If it is configured from
SAN, make sure to avoid the SAN failure points. Some of the considerations
from SAN are detailed in the following list:
– It is recommended to use a separate LUN for paging space.
– If LUN is configured through Virtual I/O Server (VIOS), consider providing
a dual VIOS with multipath (MPIO) implementation.
– It is recommended to mirror the disks at the AIX level using the mirrorvg
command.
– It is recommended to use stripped and mirrored disks.
– It is recommended to use either RAID5 or RAID10 configurations at SAN.
– Ensure FC adapter redundancy to avoid any failure.
Chapter 2. AIX configuration 49
Paging activity probable causes
The following issues are probable causes for paging activity:
› Free list reaches 0
Monitor the status of free list that AIX VMM maintains through the vmstat
command (Example 2-7 on page 45). If the free list represented as fre in the
vmstat command reaches 0 due to an outburst of memory requirement, the
system might page regardless of memory usage.
In such situations it is recommended to increase the minfree value. Always
ensure that the difference between minfree and maxfree does not exceed
1024.
› If active virtual memory exceeds the real memory, then paging happens.
Monitor the growth of active virtual memory from the vmstat output. See
Example 2-9.

Example 2-9 vmstat output showing the memory statistics

memory page
--------------------- -----------------------------------------
avm fre fi fo pi po fr sr
10560014 133369 1 114 0 0 0 0
10552156 142713 7 2106 0 0 30 38
10557399 137113 0 89 0 0 0 0
10542898 152814 7 2118 0 0 27 30
10538954 167571 0 97 0 0 188 202

Multiply the avm value by 4 K to convert from pages to bytes. From

Example 2-9, the avm value of 10538954 corresponds to 42155 MB
(10538954*4 K).
In such situation it is recommended to have more physical memory added to
the system.

Paging space related AIX commands

The following list details paging space-related AIX commands:
› Monitor the growth of paging space using lsps -a, as shown in Example 2-4.

Example 2-10 lsps -a command

# lsps -a
Page Space
Physical Volume Volume Group Size %Used Active Auto Type Chksum

hdisk0 rootvg 1152MB 21 yes yes lv 0

50 Best Practices for DB2 on AIX 6.1 for POWER Systems
 › To create additional paging space, use the mkps command. For example, to
create a paging space in volume group myvg that has four logical partitions
and is activated immediately and at all subsequent system restarts, use:
mkps -a -n -s4 myvg
› To change the attribute of the paging space use chps command. For example
to change the size of the myvg paging space use:
chps -s 4 myvg

This adds four logical partitions to the myvg paging space.

Important: It is not recommended to move the default paging space under /.

This might affect the boot sequence of the system.

2.1.6 Network tunable considerations

AIX by default provides the best recommended configurable parameters for a
network. Similar to VMM tunables, network parameters are also classified as
restricted and non-restricted tunables. These non-restricted parameters can be
modified as per the requirement but it is not recommended to modify restricted
tunables. For modifying any restricted tunables, consult IBM technical support.

From the DB2 perspective, tunables are modified for effective bandwidth use and
are modified considering the security aspect.

These tunables can be modified using the command no.

AIX no command
To display the user modifiable network tunables, use the no -a command, as
shown in Example 2-11.

Example 2-11 Output of no -a command

# no -a
arpqsize = 12
arpt_killc = 20
arptab_bsiz = 7
arptab_nb = 149
bcastping = 0

The no -F -a command can be used to display both the non-restricted and the
restricted tunables.
Chapter 2. AIX configuration 51
To modify the tunable and to make the modification permanent use the no -p -o
<tunable>=<value> command, as shown in Example 2-12.

Example 2-12 Modify and make changes permanent across reboot

# no -p -o tcp_recvspace=65536
Setting tcp_recvspace to 65536
Setting tcp_recvspace to 65536 in nextboot file
Change to tunable tcp_recvspace, will only be effective for future connections

To modify the command to be effective from the next reboot, perform the
following command, as shown in Example 2-13.

Example 2-13 Modification takes effect at the next reboot

# no -r -o ipqmaxlen=250
Setting ipqmaxlen to 250 in nextboot file
Warning: changes will take effect only at next reboot

To achieve a high level of security and performance improvement, the following

network tunable modifications are suggested for DB2 environment. The effect of
the modification varies with tunables. A few are dynamic, a few after reboot and a
few when the first connection occurs.
› Dynamic
Changes takes place immediately
› Reboot
Changes effected in the next system root
› Connect
Changes in effect when the first connection occurs.

clean_partial_connection
This parameter is recommended to be enabled to avoid SYNC attack
(unacknowledged SYN/ACK packets from server by client). As SYNC attack
leads to increases in the listen queue backlog (denial of service), removing
partial connections makes room for new non-attack connections.
› Default value: 0
› DB2 recommended value: 1
› How to set: no -p -o clean_partial_conns= 1
› Type: Dynamic

This prevents the socket listen queue to be filled up with incomplete 3-way
TCP/IP handshake partial connections
52 Best Practices for DB2 on AIX 6.1 for POWER Systems
ip6srcrouteforward
This option specifies whether or not the system forwards source-routed IPv6
packets. Source routing is a technique where the sender of a packet can specify
the route that a packet must take through the network.
› Default value: 1
› DB2 recommended value: 0
› How to set: no -p -o ip6srcrouteforward=0
› Type: Dynamic

A value of 0 causes all source-routed packets that are not at their destinations to
be discarded, preventing attacks through source route.

ipignoreredirects
This parameter is used to control ICMP Redirects. The ICMP Redirect message
is used to notify a remote host to send data packets on an alternative route.
Setting it to 1 ensures that malicious ICMP request cannot be used to create
manipulated routes.
› Default value: 0
› DB2 recommended value: 1
› How to set: no -p -o ipignoreredirects=1
› Type: Dynamic

Setting this parameter to 1 ensures that malicious ICMP request cannot be used
to create manipulated routes.

ipqmaxlen
This specifies the number of received packets that can be queued on the IP
protocol input queue. Examine ipintrq overflows using netstat -s (See
Example 2-14), and consider increasing this value as recommended.

Example 2-14 Verification of overflows

# netstat -s | grep ipintrq
0 ipintrq overflows
› Default value: 100
› DB2 recommended value: 250
› How to set: no -r -o ipqmaxlen=250
› Type: Reboot

Preventing overflows helps in processing the packets at the IP level.

Chapter 2. AIX configuration 53
An increase in value might result in more time spent in the off-level interrupt
handler, because IP has more packets to process on its input queue. This can
adversely affect processes which requires CPU time.It is best to increase
ipqmaxlen by moderate increments. See Example 2-15 on how to increase
ipqmaxlen value.
Example 2-15 ipqmaxlen
# no -r -o ipqmaxlen=250
Setting ipqmaxlen to 250 in nextboot file
Warning: changes will take effect only at next reboot

ipsendredirects
This is to specify whether the kernel must send redirect signals. Disable this
parameter to prevent illegal access through source routing.
› Default value: 1
› DB2 recommended value: 0
› How to set: no -p -o ipsendredirects=0
› Type: Dynamic

Disabling this prevents redirected packets from reaching a remote network.

ipsrcrouterecv
This parameter specifies whether the system accepts source routed packets.
Source routing is a technique where the sender of a packet can specify the route
that a packet must take through the network. The default value of 0 causes all
source-routed packets destined for this system to be discarded. A value of 1
allows source routed packets to be received.
› Default value: 0
› DB2 recommended value: 1
› How to set: no -p -o ipsrcrouterecv=1
› Type: Dynamic
This parameter ensures that the pickets destined to the destination is received.
54 Best Practices for DB2 on AIX 6.1 for POWER Systems
rfc1323
The enhancement suggested through RFC 1323 is used. RFC1323 suggests
maintaining high performance and reliability over high speed paths (bps) through
improved TCP window scale option. The TCP window scale option as defined in
RFC1323 increased the window scale size from 64 KB to 1 GB. DB2
recommends enabling this parameter for more efficient use of high bandwidth
networks.
› Default value: 0
› DB2 recommended value: 1
› How to set:
– ifconfig en0 rfc1323 1 (Interface specific, dynamic, no reboot required)
– no -p -o rfc1323=1 (system specific)
› Type: Connect

The recommended setting enables increased performance for high speed

networks. It is advisable to enable this to have higher send and receive buffer
values (greater than 64 KB)

Note: The rfc1323 network option can also be set on a per interface basis
through the ifconfig command. It can be verified using ifconfig -a.

tcp_nagle_limit
This tunable helps avoid sending large number of small packets (packet
consolidation). By default, it is set to 65535, the maximum size of IP packet in
AIX. When using DB2, we recommend that you disable it. This ensures that AIX
does not try to consolidate packets.
› Default value: 65535
› DB2 recommended value:1
› How to set: no -p -o tcp_nagle_limit=1
› Type: Dynamic

Consolidating the packets affects the real time data transfer.

TCP header size of 40bytes is sent even to send 1 byte of data across the wire,
which is a huge overhead.
Chapter 2. AIX configuration 55
tcp_nodelayack
Enabling this parameter causes TCP to send immediate acknowledgement (Ack)
packets to the sender.We recommend to set it to ON.
› Default value: 0
› DB2 recommended value: 1
› How to set: no -p -o tcp_nodelayack=1
› Type: Dynamic

Delay in sending the ACK affects the real time data transfer.

Enabling this option causes slightly more system overhead, but can result in
much higher performance for network transfers if the sender is waiting on the
receiver's acknowledgement.

tcp_recvspace
The tcp_recvspace tunable specifies how many bytes of data the receiving
system can buffer in the kernel on the receiving socket queue. The attribute must
specify a socket buffer size less than or equal to the setting of the sb_max
network attribute.
› Default value: 16384
› DB2 recommended value: 262144
› How to set:
– ifconfig en0 tcp_recvspace 262144 (interface specific, dynamic, no
reboot required)
– no -p -o tcp_recvspace=262144
› Type: Connect

With RFC1323 enabled, the recommended value ensures high data transfer rate.
Higher value might increase the workload on the adapter. Also, it decreases the
memory space for data in RAM. Monitor using vmstat for memory usage and
paging.
56 Best Practices for DB2 on AIX 6.1 for POWER Systems
tcp_sendspace
The tcp_sendspace tunable specifies how much data the sending application can
buffer in the kernel before the application is blocked on a send call.
› Default value: 16384
› DB2 recommended value: 262144
› How to set:
– ifconfig en0 tcp_sendspace 262144 (interface specific, dynamic, no
reboot required)
– no -p -o tcp_sendspace= 262144
› Type: Connect

With RFC enabled, the recommended value ensures high data transfer rate.
Higher values might increase the workload on the adapter. Also, it decreases the
memory space for data in RAM. Monitor using vmstat for memory usage and
paging.

Note:
› AIX6.1 enables use_isno by default and is a restricted variable. Because
this overrides the global network option values, it is recommended to set
the interface specific option using ifconfig for rfc1323,tcp_sendspace and
tcp_recvspace.
› The application can override all of these with the setsockopt() subroutine.

tcp_tcpsecure
This option protects TCP connections from the following three vulnerabilities:
› The first vulnerability involves the sending of a fake SYN to an established
connection to abort the connection. A tcp_tcpsecure value of 1 provides
protection from this vulnerability.
› The second vulnerability involves the sending of a fake RST to an established
connection to abort the connection. A tcp_tcpsecure value of 2 provides
protection from this vulnerability.
› The third vulnerability involves injecting fake data in an established TCP
connection. A tcp_tcpsecure value of 4 provides protection from this
vulnerability.
Chapter 2. AIX configuration 57
Values of 3, 5, 6, or 7 protects the connection from combinations of these three
vulnerabilities.
› Default value: 0
› DB2 recommended value: 5
› How to set: no -p -o tcp_tcpsecure=5
› Type: Dynamic

These settings protects TCP connections against vulnerabilities.

jumbo frames
With the advent of Gigabit Ethernet, the TCP/IP protocol now provides an ability
to send large frames (called jumbo frames). An Ethernet frame contains 1500
bytes of user data, plus its headers and trailer. By contrast, a jumbo frame
contains 9000 bytes of user data, so the percentage of overhead for the headers
and trailer is much less and data-transfer rates can be much higher.

This feature is particularly useful in environments such as replication, where

large data sets are transferred over the network. DB2 can make use of this
features by enabling it.

Maximum transmission unit (MTU)

This is the maximum size of a frame that can be sent on the wire. The MTU size
of the network can have a large impact on performance. The following list details
the recommendations:
› All devices on the same physical network, or logical network if using VLAN
tagging, must have the same MTU size.
For example, it is not recommended to have a Gigabit Ethernet adapter using
jumbo frame mode with a MTU size of 9000 bytes, while other adapters in the
network use the default MTU size of 1500 bytes.
› Configure Ethernet switches to use jumbo frames if jumbo frames are
supported on your Ethernet switch.
› With Gigabit and 10 Gigabit Ethernet, if all of the machines in the network
have Gigabit Ethernet adapters and no 10/100 adapters in the network, then it
is best to use jumbo frame mode. For example, a server-to-server connection
within the computer lab can typically be done using jumbo frames.
› It is important to select the MTU size of the adapter early in the network setup
so that all the devices and switches can be properly configured.
58 Best Practices for DB2 on AIX 6.1 for POWER Systems
 We explain the steps to enable jumbo frames on Gigbit Ethernet. Trying to
change the MTU size with the ifconfig command does not work. Use SMIT to
display the adapter settings with the following steps:
1. Select Devices.
2. Select Communications.
3. Select Adapter Type.
4. Select Change/Show Characteristics of an Ethernet Adapter.
5. Change the Transmit Jumbo Frames option from No to Yes.
The SMIT panel looks like Figure 2-5.

Figure 2-5 Enabling “jumbo frames” using smitty

Note: Jumbo frames can be used with Etherchannel. If set to Yes it forces
jumbo frames on all underlaying adapters. Virtual Ethernet adapters do not
have jumbo frames attributes, but can send jumbo sized packets.
Chapter 2. AIX configuration 59
Monitoring the network performance
You can use the following features to monitor the network performance:
› Kernel memory use by network packets:
Kernel memory buffers called mbufs are used to store data in the kernel for
incoming and outbound network traffic. Incorrect configuration affects both
network and system performance.
mbuf is controlled by:
– “thewall” network variable. No tuning is required as the system
automatically tunes the value ranging from 0-64GB. Default value is 1 GB.
– “maxmbuf” system variable defines the maximum real memory allowed for
MBUFS. If this value is “0”, it indicates that the “thewall” value is used. If
“maxmbuf >0”, then it overrides the “thewall” value. Following command
gives the “maxmbuf” value.
# lsattr -E -l sys0 | grep maxmbuf
maxmbuf 0 Maximum Kbytes of real memory allowed for MBUFS
True
– “maxmbuf” value can be changed using chdev.
chdev -l sys0 -a maxmbuf=1048576
Monitor the network memory usage using the following command.
echo "kmbucket -s" | kdb | egrep "thewall|allocated" | tr "." ""
| awk
'/thewall/ {thewall=$2} /allocated/ {allocated=$2} END{ print
"ibase=16;scale=5;",allocated,"/400/",thewall}'| bc
If the output shows .02063, it means 2.063% of mbuf has been allocated.
The threshold is indicated by sockthresh, which is 85% by default. After
the allocated memory reaches 85% of the thewall or maxmbuf value, a
new socket connection fails with ENOBUFS, until the buffer usage drops
below 85%
Monitor for any mbuf requirement failures using the netstat -m command.
See Example 2-16.

Example 2-16 netstat -m , showing the memory statistics

# netstat -m

Kernel malloc statistics:

******* CPU 0 *******

By size inuse calls failed delayed free hiwat freed
 64 539 1347842 0 6 37 1744 0
128 2098 644202 0 62 334 872 0

60 Best Practices for DB2 on AIX 6.1 for POWER Systems

256 870 187644 0 52 202 1744 0
512 2440 35547446 0 267 224 2180 0
1024 534 694480 0 129 26 872 0
2048 594 3269944 0 270 190 1308 0
4096 135 1201 0 26 20 436 0
8192 4 66 0 2 6 218 0
16384 128 134 0 18 10 109 0
32768 24 24 0 6 0 54 0
65536 59 59 0 30 0 54 0
131072 3 3 0 0 26 52 0
<rest of the cpu output is truncated for clarity>

Streams mblk statistic failures:

0 high priority mblk failures
0 medium priority mblk failures
0 low priority mblk failures

In this example:
– By size: Shows the size of the buffer.
– inuse: Shows the number of buffers of that size in use.
– failed: Shows how many allocation requests failed because no buffers
were available.

Note:
› You do not see a large number of failed calls. There might be a few, which
trigger the system to allocate more buffers as the buffer pool size
increases. If the requests for mbuf (failed) is high over a period of time,
consider increasing the thewall value.
no -o thewall=<newvalue>
› After the thewall value is increased, use vmstat to monitor total memory
use, paging activity to understand whether the increase has any negative
impact on overall memory performance.
› Kernel memory buffer (mbuf) to handle data packets are pinned kernel
memory in RAM. These memory are never paged out. Increase in mbuf
decreases the system memory for data segment.
Chapter 2. AIX configuration 61
› Monitoring the network statistics using netstat:
The netstat command can be used to determine the amount of traffic in the
network.
Use netstat -an -f inet and check columns Recv-Q and Send-Q to see
the use of the send/recvspace buffers. This also gives you the number of
connections. See Example 2-17.

Example 2-17 netstat -an -f inet

# netstat -an -f inet | more
Active Internet connections (including servers)
Proto Recv-Q Send- Local Address Foreign Address (state)
Q
tcp 0 0 *.* *.* CLOSED
tcp4 0 0 *.13 *.* LISTEN
tcp 0 0 *.21 *.* LISTEN
tcp4 0 0 *.22 *.* LISTEN
tcp 0 0 *.23 *.* LISTEN
tcp4 0 0 *.25 *.* LISTEN
tcp4 0 0 *.37 *.* LISTEN
tcp4 0 0 *.111 *.* LISTEN
tcp 0 0 *.199 *.* LISTEN
tcp 0 0 *.512 *.* LISTEN
tcp 0 0 *.513 *.* LISTEN
tcp 0 0 *.514 *.* LISTEN

Use netstat -i to understand the packets received (IPkts), the number of

input errors (Ierrs), packets transmitted (OPkts), and number of output errors
(Oerrs). See Example 2-18.

Example 2-18 netstat -i

# netstat -i
Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Coll
en0 1500 link#2 a.c.bb.1c.eb.b 27493958 0 3947570 0 0
en0 1500 9.184.65 indus65 27493958 0 3947570 0 0

If the number of errors during input packets is greater than 1% of the total
number of input packets, that is, Ierrs > 0.01 x Ipkts, run the netstat -m
command to check for a lack of memory.

Note: The nmon command can be used to understand the network statistics.
Run nmon and type n to view the details.
62 Best Practices for DB2 on AIX 6.1 for POWER Systems
2.1.7 Input and output tunable considerations
Input/output tunable parameters are managed through the ioo command.

AIX ioo command

This command sets or displays the current or next boot values for all input and
output tunables. To display the current value for all the tunable, ioo -a is used, as
shown in Example 2-19.

Example 2-19 Output of the command, ioo -a, to display the current value
# ioo -a
aio_active = 0
aio_maxreqs = 65536
aio_maxservers = 30
aio_minservers = 3
aio_server_inactivity = 300
j2_atimeUpdateSymlink = 0

Similar to VMM and network tunables, IO tunable also has a set of restricted
tunables, which can be viewed with ioo -F -a command options.

Note: Restricted tunables are not recommended to be modifed. They exist

and must be modified primarily for specialized intervention by the development
support or development teams. Contact IBM Support if any modification is
required.

To modify the tunable and to make the modification permanent use the ioo -p -o
<tunable>=<value> command. See Example 2-20.

Example 2-20 To modify and make the changes permanent

# ioo -p -o lvm_bufcnt=10
Setting lvm_bufcnt to 10 in nextboot file Setting

lvm_bufcnt to 10

To modify the tunable and to make the modification in the next boot, use the ioo
-r -o <tunable>=<value> command. See Example 2-21.

Example 2-21 To modify and make the changes after reboot

# ioo -r -o lvm_bufcnt=10
Setting lvm_bufcnt to 10 in nextboot file
Warning: changes will take effect only at next reboot
Chapter 2. AIX configuration 63
From the DB2 perspective, the default AIX6 ioo tunables are recommended.
Some of the tunables that needs attention are covered in the following sections .
Though no changes are recommended, constant monitoring using vmstat -v
and iostat -D helps to understand the impact of these parameters.

j2_maxPageReadAhead
This parameter defines the number of pages that needs to be read ahead during
the sequential file read operation on JFS2. This parameter is taken into
consideration while setting the minfree/maxfree value, as discussed in 2.1.3,
“VMM considerations for DB2” on page 40.
› Default value: 128
› DB2 recommended value: 128

j2_maxRandomWrite
This parameter specifies a threshold for random writes to accumulate in RAM
before subsequent pages are flushed to disk by the JFS2's write-behind
algorithm.The random write-behind threshold is on a per-file basis.

For DB2, we recommend that you use the default value. The default value of “0”
disables random write-behind and indicates that random writes stay in RAM until
a sync operation. If vmstat n shows page out and I/O wait peaks on regular
intervals (usually when the sync daemon is writing pages to disk), useful to set
this value to 1 or higher if too much I/O occurs when syncd runs.
› Default value: 0
› DB2 recommended value: 0
› How to set: ioo -p -o j2_maxRandomWrite=<value>
› Type: Dynamic

j2_minPageReadAhead
This parameter specifies the minimum number of pages to be read ahead when
processing a sequentially accessed file on Enhanced JFS. This is useful for
sequential access. A value of 0 might be useful if the I/O pattern is purely
random. For DB2, we recommend that you use the default value, which is 2.
› Default value: 2
› DB2 recommended value: 2

maxpgahead
This parameter specifies the maximum number of pages to be read ahead when
processing a sequentially accessed file. This is a restricted tunable in AIX 6.1.
For DB2, we recommend that you use the default value, which is 8.
› Default value: 8
› DB2 recommended value: 8
64 Best Practices for DB2 on AIX 6.1 for POWER Systems
minpgahead
This parameter specifies the number of pages with which sequential read-ahead
starts. This is a restricted tunable in AIX6.1. For DB2, we recommend that you
use the default value, which is 2.
› Default value: 2
› DB2 recommended value: 2

Tip: Set both the maxnpgahead and minpgaheadparameters to a power of

two. For example, 2, 4, 8,...512, 1042... and so on.

maxrandwrt
This parameter specifies a threshold (in 4 KB pages) for random writes to
accumulate in RAM before subsequent pages are flushed to disk by the
write-behind algorithm. A value of 0 disables random write-behind and indicates
that random writes stay in RAM until a sync operation. Setting maxrandwrt
ensures these writes get flushed to disk before the sync operation has to occur.
› Default value: 0
› DB2 recommended value: 0
› How to set: ioo -p -o maxrandwrt=<threshold value>
› Type: Dynamic

If vmstat n shows page out and I/O wait peaks on regular intervals (usually when
the sync daemon is writing pages to disk), set it to a threshold level to reduce the
high I/O. However, this can degrade performance because the file is being
flushed each time after the threshold value is reached. Tune this option to favor
interactive response time over throughput. After the threshold is reached, all
subsequent pages are immediately flushed to disk.
j2_nBufferPerPagerDevice
Specifies the number of file system bufstructs for JFS2.This is a restricted
tunable. For DB2, we recommend that you use the default value, which is 512.
› Default value: 512
› DB2 recommended value: 512
numfsbufs
This parameter specifies the number of file system bufstructs. This is a restricted
tunable. For DB2, we recommend that you use the default value, which is 196.
› Default value: 196
› DB2 recommended value: 196

Note: If you make any changes to the file system bufstructs tunables
(j2_nBufferPerPagerDevice and numfsbufs), the new values take effect only
when the file system is remounted.
Chapter 2. AIX configuration 65
j2_nPagesPerWriteBehindCluster
JFS file systems and JFS2 file systems are partitioned into 16 KB partitions or 4
pages. Each of these partitions is called a cluster. In JFS2, this parameter
specifies the number of pages per cluster processed by JFS2's write behind
algorithm.
› Default value: 32
› DB2 recommended value: 32

This is useful to increase if there is a need to keep more pages in RAM before
scheduling them for I/O when the I/O pattern is sequential. It might be
appropriate to increase if striped logical volumes or disk arrays are being used.

pv_min_pbuf
This specifies the minimum number of pbufs per PV that the LVM uses.This is a
global value that applies to all VGs on the system. The lvmo command can be
used to set a value for a particular VG. In this case, the higher of the two values is
used for this particular VG.
› AIX default: 512
› DB2 recommended value: 512

Note: This is a restricted tunable in AIX. The best practise is to monitor

"pending disk I/Os blocked with no pbuf" in vmstat -v and discuss with IBM
technical support to consider any further modification.

sync_release_ilock
This parameter specifies whether the i-node lock is to be held or not while
flushing I/O to file when the sync daemon is running.

If set, flush all I/O to a file without holding the i-node lock, and use the i-node lock
to do the commit.

A value of 0 indicates off and that the i-node lock is held while all dirty pages of a
file are flushed. I/O to a file is blocked when the syncd daemon is running.

This is a restricted tunable in AIX. For DB2, we recommend that you use the
default value, which is 0.
› Default value: 0
› DB2 recommended value: 0

Note: The j2_ parameters are not applicable for DB2 table spaces configured
with the FILE SYSTEM CACHING DB2 parameter, where the file system
caching is not involved.
66 Best Practices for DB2 on AIX 6.1 for POWER Systems
Asynchronous IO (AIO) consideration for DB2
Asynchronous Input Output (AIO) is a software subsystem within AIX that allows
a process to issue an I/O operation and continue processing without waiting for
the I/O to finish.
Asynchronous I/O operations run in the background and do not block user
applications. This improves performance because I/O operations and
applications processing run simultaneously. Applications such as databases and
file servers take advantage of the ability to overlap processing and I/O.

There are two AIO subsystems:

› Legacy AIO
› Posix AIO

The difference between them is parameter passing. Both the subsystem can run
concurrently in the system.
From AIX6.1 onwards, no AIO servers are started by default. They are started
when applications initiate AIO requests and stay active as long as they are
servicing requests.
› When AIO is enabled, minimum number of AIO servers are created based on
the minserver value.
› If additional servers are required, more AIO servers are added up to the
maximum, based on the maxserver value.
› The maximum number of outstanding requests is defined by maxreqs.
› These are all ioo tunables. Changes do not require system reboot.

Note: In AIX6 all AIO parameters become the ioo command tunables. The
aioo command used in the previous version of AIX5.3 is removed.

From the DB2 perspective, because the AIO tuning is dynamic, no tuning is
required.
› Default AIO settings of minserver, maxserver, maxreqs are recommended. No
change is required.
› AIO fast path (aio_fastpath) is also enabled by default, so no changes are
required. This is a restricted tunable, as well.
› AIO CIO fastpath (aio_fsfastpath) is also enabled by default and so, no
changes are required. This is a restricted tunable, as well.

Note: AIO can be used with file systems mounted with DIO/CIO. DB2 uses
DIO/CIO when the table space is used with NO FILE SYSTEM CACHING.
Chapter 2. AIX configuration 67
AIO monitoring and tuning suggestions
The following list describes AIO monitoring and tuning suggestions:
› To understand the number of active AIO servers, use the pstat -a | grep -c
aioscan command.
› To get the general AIO report, use the iostat command:
iostat -AQ 1 5
This command displays the output for every second. Monitoring has to be
done over a period of time to understand the status of aio server.
› To understand the state of the each active AIO servers, use the following
command:
ps -vg | egrep “aio | SIZE”
From the output in Example 2-22, we can identify the %CPU,%MEM and size
consumed by the each aio kernel process.

Example 2-22 ps vg | egrep "aio|SIZE"

# ps vg | egrep "aio|SIZE"
PID TTY STAT TIME PGIN SIZE RSS LIM TSIZ TRS %CPU %MEM COMMAND
127048 -A 0:00 9 448 448 xx 0 0 0.0 0.0 aioserver
385024 -A 0:00 8 448 448 xx 0 0 0.0 0.0 aioserver
442584 -A 0:00 9 448 448 xx 0 0 0.0 0.0 aioserver
446682 -A 0:00 11 448 448 xx 0 0 0.0 0.0 aioserver
450780 -A 0:00 7 448 448 xx 0 0 0.0 0.0 aioserver
454878 -A 0:00 8 448 448 xx 0 0 0.0 0.0 aioserver
458976 -A 0:00 8 448 448 xx 0 0 0.0 0.0 aioserver
463074 -A 0:00 7 448 448 xx 0 0 0.0 0.0 aioserver
467172 -A 0:00 7 448 448 xx 0 0 0.0 0.0 aioserver
471270 -A 0:00 5 448 448 xx 0 0 0.0 0.0 aioserver
475368 -A 0:00 6 448 448 xx 0 0 0.0 0.0 aioserver
479466 -A 0:00 6 448 448 xx 0 0 0.0 0.0 aioserver
483564 -A 0:00 2 448 448 xx 0 0 0.0 0.0 aioserver
487662 -A 0:00 0 448 448 xx 0 0 0.0 0.0 aioserver

› Tuning consideration for the aio_maxservers is as follows. Monitor the status

over the period of time using these commands.
– If all the active aio servers are consuming higher CPU%, then the
maxservers value can be increased by 10%. For example, increase the
default value of 30 to 33.
– If a few of the aio servers are consuming less CPU, the system has the
required server as it needs.
– To modify the “aio_maxservers” value, following option has to be used ioo
-o aio_maxservers=<value>. It does not require a reboot. See
Example 2-23 on page 69 for an example on how to modify the
aio-maxservers value.
68 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Example 2-23 changing aio_maxservers value

# ioo -o aio_maxservers=33
Setting aio_maxservers to 33 #
ioo -a | grep aio
aio_active = 0
aio_maxreqs = 65536
aio_maxservers = 33
aio_minservers = 3
aio_server_inactivity = 300

Note: AIX6 introduces a new parameter, aio_server_inactivity, that controls

how long in seconds the AIO server sleeps waiting for work. The main benefit
at the AIX layer is to free pinned memory and decrease the number of
processes after a peak workload activity with the AIO subsystem, which helps
lighten the load process scheduling and reduce system resource usage.

The default value is 300 seconds.

In summary, for DB2 we do not recommend any changes to the default ioo
tunable values in AIX 6.1. Any further modification is to be considered based on
the monitoring over a period of time as described.

2.1.8 Scheduler tunable considerations

Processor scheduler tunable parameters are managed with the command
schedo. Similar to other tunables such as VMM, IO and network, scheduler
tunables are also classified into both restricted and non-restricted tunables.

In general, we do not recommend that you make any changes with respected to
the scheduler tunables for DB2 environments. From the AIX tunable perspective,
apart from the recommended modifications, if a performance degradation is still
observed, relevant performance data is submitted to IBM for analysis.

Note: There might be specific cases where changing the restricted values
might improve the performance. In this situation, contact technical support for
further analysis. It is never recommended to change the restricted tunables
from their defaults.
Chapter 2. AIX configuration 69
2.1.9 DB2 Groups, users, and password configuration on AIX
You can create required groups, users and set initial password using the
commands in Example 2-24.

Example 2-24 Create required groups, users and set initial password
mkgroup id=999
db2iadm1 mkgroup
id=998 db2fadm1
mkgroup id=997 dasadm1
mkuser id=1004 pgrp=db2iadm1 groups=db2iadm1 home=/db2home/db2inst1 db2inst1
mkuser id=1003 pgrp=db2fadm1 groups=db2fadm1 home=/db2home/db2fenc1
db2fenc1
mkuser id=1002 pgrp=dasadm1 groups=dasadm1 home=/db2home/dasusr1
dasusr1 passwd db2inst1
passwd db2fenc1

passwd dasusr1

Note: To enable long password support on the AIX 6.1, install APAR IZ35001.

2.1.10 DB2 user ID resource limits (ulimits)

For DB2 instance owner, fenced user, and administration user IDs set user
process resource limits (ulimits) to unlimited using the chuser command, as
shown in Example 2-25.

Example 2-25 chuser command

chuser fsize=-1 fsize_hard=-1 data=-1 data_hard=-1 stack=-1 stack_hard=-1 rss=-1

rss_hard=-1 nofiles=-1 nofiles_hard=-1 db2inst1

2.1.11 File system mount point permissions

DB2 installation in a separate file system:
› For root installations when DB2 product is installed on separate file system or
on a non-default directory, ensure others are given read and execute
permission for underlying mount point.
For example, when DB2 is installed in /db2bin/opt/IBM/db2/V9.7/ and
 /db2bin is a separate file system, ensure others have read and execute file
permissions on empty /db2bin mount point directory
This allows non-root users to successfully execute the db2ls command. Using
db2ls, you can list installed DB2 products and features.

70 Best Practices for DB2 on AIX 6.1 for POWER Systems

 DB2 instance directory in a separate file system:
In scenarios where a separate file system is used for DB2 instance home
directory, ensure that DB2 instance ID owns the mount directory (home dir) for
successful DB2 instance creation.

For example db2inst1 must own the /db2home/db2inst1directory when a

separate file system is mounted on it.

2.1.12 File system mount options

The following list details the file system mount options:
› Mount option rbrw is recommended for specific file systems where FILE
SYSTEM CACHING is used to get better performance (such as temporary
table spaces, table spaces with LONG and LOB data, BACKUP file systems).
The option rbrw specifies that when sequential read or sequential write of a
file in this file system is detected, the real memory pages used by the file are
released after the pages are written or read to disk. This helps in improving
the performance of syncd, which flushes the real memory pages to disk.
› Do not use explicit dio/cio mount options for file systems, which are used for
DMS table spaces defined with NO FILE SYSTEM CACHING as DB2
chooses the best I/O method required.

2.1.13 Maximum number of AIX processes allowed per user:

DB2 users (such as the fenced user ID) can create all the required processes.
The maxuproc kernel parameter helps in limiting the number of process that can
be created by a user. DB2 recommends the value of 4096.

Verify the existing “maxuproc” value as follows:

lsattr -E -l sys0 -a maxuproc

Monitor the maximum number of processes under DB2 instance ID using the ps
–fu db2inst1 or db2_local_ps command and adjust the maxuproc value
accordingly. The following example sets the maxuproc to 4096
chdev –l sys0 –a maxuproc=4096

Reboot is required to make the changes effective.

Chapter 2. AIX configuration 71
2.2 DB2 registry variables
In the following sections we discuss DB2 registry variables. The registry variables
can be examined or modified using the db2set command.

DB2_LOGGER_NON_BUFFERED_IO
› Default value: AUTO
› Recommended: Default

See ‘3.2.2, “Tablespace design” on page 92’ to read more about buffered and
non-buffered IO.

Starting with Version 9.7, the default value for this variable is AUTOMATIC. When
set to AUTOMATIC, active log files are opened with DIO. This eliminates the
operating system overhead of caching database recovery logs. The database
manager determines which log files benefit from using non-buffered I/O.

When set to ON, all active log files are always opened with DIO. When set to
OFF all active log files are buffered I/O. In Version 9.5 Fix Pack 1 or later, the
default was OFF.

Do not use any explicit mount options for the file systems used for DB2 log files.

DB2_USE_IOCP
› Default value: ON
› Recommended: Default

IOCP has to be configured before enabling this variable. This feature enables the
use of AIX I/O completion ports (IOCP) to submit and collect asynchronous I/O
(AIO) requests and enhance performance in a non-uniform memory access
(NUMA) environment by avoiding remote memory access. This is also available
on DB2 v9.5 starting from fixpack 3.

It is recommended to leave this parameter at its default. You might monitor the
system IO statistics or nmon to fine-tune this parameter.

Note: DB2_USE_IOCP is ON by default since DB2 v9.7.

72 Best Practices for DB2 on AIX 6.1 for POWER Systems
 DB2_USE_FAST_PREALLOCATION
› Default value: ON
› Recommended: Default

This feature is to reserve table space and speed up the process of creating or
altering large table spaces. This is available from DB2 v9.7 fixpack 1 and DB2
v9.5 fixpack 5.

DB2_PARALLEL_IO
› Default value: NULL
› Recommended: Refer to Storage chapter to set an optimum value for this
registry variable.

2.2.1 DB2_Resource_policy
In the following sections we discuss more advanced DB2 registry variables.

DB2_RESOURCE_POLICY
› Default value:
› Recommended:

This variable defines a resource policy that can be used to limit what operating
system resources are used by the DB2 database. It can contains rules for
assigning specific operating system resources to specific DB2 database objects.

This registry variable can be used to limit the set of processors that the DB2
database system uses. The extent of resource control varies depending on the
operating system.

On AIX NUMA and Linux NUMA-enabled machines, a policy can be defined that
specifies what resource sets the DB2 database system uses. When resource set
binding is used, each individual DB2 process is bound to a particular resource
set. This can be beneficial in performance tuning scenarios.
Chapter 2. AIX configuration 73
The following steps illustrate AIX resource sets configuration to enable processor
affinity for DB2 partitions. A single node with 16 processors and eight logical
database partitions is considered.
1. Define AIX resource sets in /etc/rsets file.

Define eight new resource sets in the /etc/rsets files to use CPU Number 8
through 15. These 8eight resource sets are named as DB2/MLN[1-8], as
shown in Example 2-26.

Example 2-26 Resource sets defined in /etc/rsets file

DB2/MLN1:
owner = db2tpch
group =
db2admin perm =
rwr-r-
resources = sys/cpu.00008
DB2/MLN2:
owner = db2tpch
group =
db2admin perm =
rwr-r-
resources = sys/cpu.00009
DB2/MLN3:
owner = db2tpch
group =
db2admin perm =
rwr-r-
resources = sys/cpu.00010
DB2/MLN4:
owner = db2tpch
group =
db2admin perm =
rwr-r-
resources = sys/cpu.00011
DB2/MLN5:
owner = db2tpch
group =
db2admin perm =
 rwr-r-
resources = sys/cpu.00012
DB2/MLN6:
owner = db2tpch
group =
db2admin perm =
rwr-r-
resources = sys/cpu.00013
DB2/MLN7:
owner = db2tpch
group =
db2admin perm =
rwr-r-
resources = sys/cpu.00014

74 Best Practices for DB2 on AIX 6.1 for POWER Systems

 DB2/MLN8:
owner = db2tpch
group =
db2admin perm =
rwr-r-
resources = sys/cpu.00015

2. Update the AIX kernel.

The newly defined resource sets are added to the kernel data structures
using the following SMIT fast path:
$ smit reloadrsetcntl
This menu gives the option to reload the database now, at next boot, or at
both times. Select both such that the rset loads now and after each reboot.

3. Update db2nodes.cfg.

The db2nodes.cfg file resides in INSTANCE_OWNER/sqllib directory. The file

format is:
nodenum hostname logical_port netname resourcesetname
After updating the eight resource names in the resourcesetname field,
db2nodes.cfg looks like Example 2-27.

Example 2-27 Resource names in db2nodes.cfg file

1 CLYDE 0 CLYDE DB2/MLN1
2 CLYDE 1 CLYDE DB2/MLN2
3 CLYDE 2 CLYDE DB2/MLN3
4 CLYDE 3 CLYDE DB2/MLN4
5 CLYDE 4 CLYDE DB2/MLN5
6 CLYDE 5 CLYDE DB2/MLN6
7 CLYDE 6 CLYDE DB2/MLN7
8 CLYDE 7 CLYDE DB2/MLN8

2.2.2 DB2 memory registry variables

Following memory related registry parameters only applicable when STMM is
enabled.

Use the nmon, vmstat, lsps commands to monitor and fine-tune the following
memory-related registry parameters. Refer to 2.1.3, “VMM considerations for
DB2” on page 40 for more information about monitoring and tuning VMM
parameters.
Chapter 2. AIX configuration 75
DB2_LARGE_PAGE_MEM
› Default value: NULL
› Recommended: Refer to 2.1.4, “Large page considerations” on page 46

Use this registry variable to improve performance for memory access-intensive

applications that use large amounts of virtual memory.

To enable the DB2 database system to use them, the operating system must be
configured to use large or huge pages. Refer to 2.1.4, “Large page
considerations” on page 46 for all pre-requisites and steps to set this parameter.

DB2MEMDISCLAIM
› Default value: YES
› Recommended: Default

Memory used by DB2 database system processes might have associated paging
space. This paging space might remain reserved even when the associated
memory has been freed. Whether or not this is much depends on the operating
system's (tunable) virtual memory management allocation policy.
The DB2MEMDISCLAIM registry variable controls whether DB2 agents explicitly
requests that the operating system disassociate the reserved paging space from
the freed memory.

A setting of Yes results in smaller paging space requirements and less IO might
provide performance benefit. However, when there is plenty of real memory and
paging space, a setting of No might yield performance benefit.

You need to monitor the paging space use and VMM parameters (see 2.1.3,
“VMM considerations for DB2” on page 40) and set this registry value
accordingly.

DB2_MEM_TUNING_RANGE
› Default value: NULL
› Recommended: Default

Using this registry parameter, DB2 instance sets minimum and maximum free
physical memory thresholds available to the server (for other applications). It is
recommended to leave at default. The setting of this variable has no effect unless
the self-tuning memory manager (STMM) is enabled and database_memory is
set to AUTOMATIC.

You need to monitor the paging space use and other VMM parameters (see
2.1.3, “VMM considerations for DB2” on page 40) to adjust the thresholds
76 Best Practices for DB2 on AIX 6.1 for POWER Systems
2.2.3 DB2 Communications registry variables
Use nmon - Network Interface View (‘n’) to monitor network performance and
fine-tune this parameter. You might also use netstat and entstat to monitor and
fine-tune any of the following TCP/IP registry parameters. See 2.1.6, “Network
tunable considerations” on page 51.

Note: Following registry parameters override OS level settings.

DB2_FORCE_NLS_CACHE
› Default value: FALSE
› Recommended: Default

This variable is used to eliminate the chance of lock contention in multi-threaded

applications. When this registry variable is TRUE, the code page and territory
code information is saved the first time a thread accesses it. From that point, the
cached information is used for any other thread that requests this information.

This eliminates lock contention and results in a performance benefit in certain

situations. This setting must not be used if the application changes locale
settings between connections. It is probably not needed in such a situation
because multi-threaded applications typically do not change their locale settings
because it is not thread safe to do so.

DB2TCPCONNMGRS
› Default value: 1
› Recommended:Default

This variable determines the number of TCP/IP connection managers on a given

server. With multiple TCP/IP connection managers there is a performance boost
on remote connections in systems with a lot of users, frequent connects, and
disconnects, or both.

Note: Each additional TCP/IP connection manager has additional overhead to

memory and storage.

Use nmon - Network Interface View (‘n’) to monitor network performance and
fine-tune this parameter.
Chapter 2. AIX configuration 77
DB2SORCVBUF and DB2SOSNDBUF
› Default value: 65536
› Recommended: Default

The DB2SORCVBUF registry variable determines the value of TCP/IP receive

buffers. DB2SOSNDBUF determines the value of TCP/IP send buffers.

Use nmon - Network Interface View (‘n’) to monitor network performance and
fine-tune this parameter. You might also use /usr/bin/entstat –d <interface
name> to display all the statistics.

DB2_HADR_SORCVBUF and DB2_HADR_SOSNDBUF

› Default value: Set by OS
› Recommended: Default

To maximize network and HADR performance, the TCP socket buffer sizes might
require tuning. If you change the TCP socket buffer size at the system level, the
settings are applied to all TCP connections on the machine. Setting a large
system level socket buffer size consumes a large amount of memory.

These two registry variables allow tuning of the TCP socket send and receive
buffer size for HADR connections only. They have the value range of 1024 to
4294967295 and default to the socket buffer size of the operating system, which
varies depending on the operating system. Some operating systems
automatically round or silently cap the user-specified value.

Note: HADR log shipping workload, network bandwidth, and transmission

delay are important factors to consider when tuning the TCP socket buffer
sizes. These factors can be monitored using nmon - Network Interface View
(‘n’).

DB2CHECKCLIENTINTERVAL
› Default value: 50
› Recommended: Default

This variable specifies the frequency of TCP/IP client connection verifications.

For heavy DB2 workloads, setting this parameter to a high value (less frequent
verifications) might provide performance benefit.
78 Best Practices for DB2 on AIX 6.1 for POWER Systems
DB2TCP_CLIENT_CONTIMEOUT
› Default value: 0
› Recommended: Default

The DB2TCP_CLIENT_CONTIMEOUT registry variable specifies the number of

seconds a client waits for the completion on a TCP/IP connect operation. If a
connection is not established in the seconds specified, the DB2 database
manager returns the error -30081 selectForConnectTimeout.

There is no timeout if the registry variable is not set or is set to 0.

Note: Operating systems also have a connection timeout value that might take
effect prior to the timeout you set using DB2TCP_CLIENT_CONTIMEOUT.
For example, AIX has a default tcp_keepinit=150 (in half seconds) that
terminates the connection after 75 seconds.

DB2TCP_CLIENT_KEEPALIVE_TIMEOUT
› Default value: 0
› Recommended: Default

The DB2TCP_CLIENT_KEEPALIVE_TIMEOUT registry variable specifies the

maximum time in seconds before an unresponsive connection is detected as no
longer alive. When this variable is not set, the system default TCP/IP keep alive
setting is used (typically two hours). Setting this parameter to a lower value than
the system default allows the database manager to detect connection failures
sooner, and avoids the need to reconfigure the system default, which impacts all
TCP/IP traffic and not connections established by DB2.

DB2TCP_CLIENT_RCVTIMEOUT
› Default value: 0
› Recommended: Default

The DB2TCP_CLIENT_RCVTIMEOUT registry variable specifies the number of

seconds a client waits for data on a TCP•IP receive operation. If data from the
server is not received in the seconds specified, then the DB2 database manager
returns the error -30081 selectForRecvTimeout.

There is no timeout if the registry variable is not set or is set to 0.

Note: The value of the DB2TCP_CLIENT_RCVTIMEOUT can be overridden

by the CLI, using the db2cli.ini keyword ReceiveTimeout or the connection
attribute SQL_ATTR_RECEIVE_TIMEOUT
Chapter 2. AIX configuration 79
2.2.4 DB2 Database Manager Configuration (DBM) parameters
There are many DB2 Database Manager Configuration (DBM) parameters that
have to be set specific to each workload. On AIX DLPARs set CPUSPEED,
COMM_BANDWIDTH to -1 such that they are computed appropriately by DB2.

2.2.5 DB2 Database Configuration (DB) parameters

Starting with DB2 v9.1, Self Tuning Memory Manager (STMM) feature has been
added, which simplifies memory management by setting critical memory related
parameters to AUTOMATIC.

Note: In addition to setting STMM to AUTOMATIC, the database

configurations for the memory consumers are set to AUTOMATIC. These
consumers include:
› bufferpool
› lock
› package cache
› sort heap

2.3 Configurational differences: AIX 5.3 versus AIX 6.1

Table 2-1 summarizes the configurational differences between AIX 5.3 and AIX
6.1.

Table 2-1 Tunables: AIX 5.3 versus AIX 6.1

Parameter Parameter AIX 5.3 AIX 6.1 DB2 best Comments
group default default practice
setting setting

ioo j2_maxPageRead 128 128 128 The vmo tunable minfree is

Ahead at least set to this value.

ioo j2_maxRandomWrite 128 0 0 Random write behind is

disabled hence the file
pages flushed to disk only
during sync.

ioo j2_minPageRead 2 2 2 No change in value from AIX

Ahead 5.3.

ioo j2_nBufferPerPager 512 512 512 Restricted tunable in AIX

Device 6.1.
80 Best Practices for DB2 on AIX 6.1 for POWER Systems
Parameter Parameter AIX 5.3 AIX 6.1 DB2 best Comments
group default default practice
setting setting

ioo maxpgahead 128 8 8 Restricted tunable in AIX 6.1

ioo maxrandwrt 128 0 0 Restricted tunable in AIX

6.1.Similar to JFS Random
write.

ioo minpgahead 2 2 2 No change in value from

AIX 5.3.

ioo numfsbufs 4096 196 196 Restricted tunable in AIX 6.1

ioo sync_release_ilock 1 0 0 Restricted tunable in AIX

6.1.

no bcastping 0 0 0 No change in value from AIX

5.3.

no clean_partial_conns 1 0 1 Enabled to overcome SYNC

attack.

no directed_broadcast 0 0 0 No change in value from AIX

5.3.

no extendednetstats 0 0 Restricted tunable in AIX

6.1.

no icmpaddresmask 0 0 0 Disable Internet Control

Message Protocol (ICMP)
query message. No change
in value from AIX 5.3.

no ip6srcrouteforward 0 1 0 IPv6 source forwarding is

disabled.

no ipforwarding 0 0 0 No change from AIX 5.3

recommendation.Kernel
does not involve in IP
forwarding.

no ipignoreredirects 1 0 1 Used to control ICMP

redirects and setting it to 1
ensures that malicious
ICMP request cannot be
used to create manipulated
routes.
Chapter 2. AIX configuration 81
Parameter Parameter AIX 5.3 AIX 6.1 DB2 best Comments
group default default practice
setting setting

no ipqmaxlen1 250 100 250 Default will not handle large

workloads. Hence increased
to 250.

no ipsendredirects 0 1 0 Disabled to prevent illegal

access through source
routing.

no ipsrcrouteforward 1 1 1 Source route forwarding is

encouraged.

no ipsrcrouterecv 0 0 1 We recommend 1 as
required by topsvcs.

no ipsrcroutesend 1 1 1 No change in value from AIX

5.3.

no nonlocsrcroute 0 1 1 Source route encouraged

no rfc1323 1 0 1 The rfc1323 tunable enables

the TCP window scaling
option.

no sb_max 1310720 104857 1048576 Number of socket buffer

6 queued with each sockets.
More socket buffer space
provided compared with AIX
5.3.

no tcp_mssdflt 1460 1460 Segment size used in

communicating with remote
networks.

no tcp_nagle_limit 1 65535 1 DB2 already disable this per

connection. This tunable will
ensure AIX does not try to
consolidate packets.

no tcp_nodelayack 1 0 1 Does not want to delay the

ACK.

no tcp_pmtu_discover 0 1 0 We recommend 0 to prevent

illegal access through
source routing.
82 Best Practices for DB2 on AIX 6.1 for POWER Systems
Parameter Parameter AIX 5.3 AIX 6.1 DB2 best Comments
group default default practice
setting setting

no tcp_recvspace 262144 16384 262144 The tcp_recvspace tunable

specifies how many bytes of
data the receiving system
can buffer in the kernel on
the receiving sockets queue.
This can also be set through
DB2 overrides.

no tcp_sendspace 262144 16384 262144 The tcp_sendspace tunable

specifies how much data the
sending application can
buffer in the kernel before
the application is blocked on
a send call. This can also be
set through DB2 overrides.

no tcp_tcpsecure3 0 5

no use_isno4 0 1 1 Restricted tunable in AIX

6.1.

vmo cpu_scale_memp 8 8 8 Restricted tunable in AIX

6.1.

vmo page_steal_method 0 1 1 Restricted tunable in AIX6.1

vmo lru_file_repage 1p 0 0 Restricted tunable in AIX

6.1.

vmo maxclient% 80 90 90 Restricted tunable in AIX

6.1.

vmo maxfree minfree+ 1088 minfree+ Difference between minfree

512 512 and maxfree must not be
more than 1000.

vmo maxperm% 80 90 90 Restricted tunable in AIX

6.1.

vmo memory_affinity 1 1 1 Restricted tunable in AIX

6.1.
Chapter 2. AIX configuration 83
Parameter Parameter AIX 5.3 AIX 6.1 DB2 best Comments
group default default practice
setting setting

vmo minfree 4096 960 4096 It is at least the size of

[memory minimum page read ahead
less than value.
8 GB]

8192
[memory
greater
than 8
GB

vmo minperm% 20 3 3 Set to low value to avoid

computational pages to be
paged.

vmo strict_maxclient 0 1 1 Restricted tunable in AIX

6.1.
84 Best Practices for DB2 on AIX 6.1 for POWER Systems
 3

Chapter 3. Storage layout

This chapter describes the storage principles guiding the storage layout, the
mapping of physical storage to logical storage, the layout of file systems, and
how DB2 uses them.

This chapter has the following sections:

› “Introduction” on page 86
› “DB2 storage design” on page 87
› “Storage hardware” on page 100
› “Tuning storage on AIX” on page 115
› “LVM configuration: Volume groups and logical volumes” on page 119
› “File systems” on page 126
› “Conclusion” on page 128
© Copyright IBM Corp. 2010. All rights reserved. 85
3.1 Introduction
The main objective of a database is to store and retrieve data quickly and
efficiently. One of the most important considerations when designing a database
is its storage layout. Where the data is placed can significantly affect the
performance of the operations against the database, and when designed and
implemented it can be hard to change. Therefore, it is important for database
administrators (DBAs) to understand the advantages and disadvantages of
choosing a particular strategy. Furthermore, understanding the concepts related
to the physical layout is essential when DBAs have to discuss with the system
administrator how the storage needs to be configured

In this chapter, we refer to two scenarios: one for OLTP environments and one for
data warehouse environments, as each has its own requirements. Figure 3-1 and
Figure 3-2 on page 87 show a high level graphical view of a typical storage layout
in both environments.

Typical OLTP DB2 Storage Layout

Figure 3-1 Typical OLTP storage layout

86 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Typical Data Warehouse DB2 Storage Layout

Figure 3-2 Typical Data Warehouse storage layout

In the rest of this chapter we discuss the details on how to build and configure
such environments for best performance.

3.2 DB2 storage design

The first step in designing a database storage layout is to understand what
storage space is needed. In a DB2 database, storage is required for the following
data categories:
› Instance directory
› Catalog table and database directory
› Permanent data: Tables and indexes
› Temporary data
› Transaction logs
› Backup data and archived logs
Chapter 3. Storage layout 87
 One simple but effective design is to store each of these data categories in a
dedicated file system. These file systems are resided on separate physical disks.
Further discussion on the mapping between file systems and disk spindles is
presented in subsequent sections.

This simple design ensures that categories of data are being accessed without
interfering with others. As a result, the database system can achieve better
performance and availability. In an OLTP (online transaction processing)
environment, in particular, separating transaction logs from the
permanent/temporary data is important because of its high transaction rate. For
a DW (data warehouse) environment, however, permanent data and transaction
logs might be collocated in the same physical disks. This is because DW
workloads are often read-only and do not have as high transaction rate as OLTP
workloads. Thus, the demand on transaction logs is lower. Allowing the sharing
of disks gives DBAs the flexibility to manage storage space and better use any
unused space.

On the other hand, this simple design can help with problem determination. For
example, if we observe that a set of disks corresponding to a file system is much
busier than the others, we can identify which part of the system or data is hot.
Better parallelism or more resources (for example, more number of spindles)
might be needed to alleviate the issue.

3.2.1 Automatic storage

To make storage management easier, starting from V8.2 FP9, DB2 introduces a
feature called automatic storage. This feature allows storage to be managed at
the database level. The responsibility of creating, extending, and adding
containers is taken over by the database manager. We no longer need to
explicitly define containers for the table space.

In DB2 v9.7, all databases are created with automatic storage1 by default. In
particular, when creating a database, we establish one or more initial storage
paths in which table spaces have their containers. For example:
CREATE DATABSE TESTDB on /path1, /path2

This command creates the database with two storage paths: /path1 and /path2.
The recommendation is for each storage path to reside in a separate file system.
With the CREATE DATABASE command, the default table spaces created by DB2
(for example, SYSCATSPACE, USERSPACE1, and TEMPSPACE) now have two
containers, one on each of the two storage paths.

1 Ifyou do not want to use automatic storage for a database, you must explicitly specify the
AUTOMATIC STORAGE NO clause on the CREATE DATABASE command.
88 Best Practices for DB2 on AIX 6.1 for POWER Systems
Any new table spaces created without explicitly providing containers definitions
are also created in these two storage paths. For example:
CREATE TABLESPACE TBSP1

This creates a table space with two containers, one on each of the two storage
paths (/path1 and /path2). As the database grows, the database manager
automatically extends the size of containers across all the storage paths.

Furthermore, automatic storage allows us to add a new storage space if we run

out of space in the existing storage paths. For example, suppose we want to add
extra storage space to the database from the previous example, and this storage
space is in a new storage path, say /path3. We use the ALTER DATABASE
statement as follows:
ALTER DATABASE TESTDB ADD STORAGE ON /path3

The new storage path is not used until there is no more room to grow within the
containers on the existing storage paths (/path1 and /path2). To use the new
storage path immediately, issue an ALTER TABLESPACE statement to rebalance all
the database table spaces resided in the existing storage paths. For instance:
ALTER TABLESPACE TBSP1 REBALANCE

The rebalance process runs asychronously in the background and does not
affect the availability of data. However, rebalance is an expensive operation and
has significant IO overhead. Therefore, it is important to gauge the performance
impact on the database system when rebalancing is being performed.

One suggestion is to start with rebalancing a relatively small table space and use
the result as a reference to estimate how long it takes to rebalance the other
table spaces, and what the performance impact is. Then, decide the time of day
that is suitable for performing the rebalance of larger table spaces.

Alternatively, we can use the throttling utility in DB2 to limit the performance
impact of rebalancing on the system. The database manager configuration
parameter util_impact_lim sets the limit on the impact of which all throttled
utilities can have on the overall workload of the system. By default, this
parameter is set to 10, which means all throttled utilities combined can have no
more than a 10% average impact upon the workload as judged by the throttling
algorithm. A value of 100 indicates no throttling. The SET UTIL_IMPACT_PRIORITY
command is used to set the priority that a particular utility has over the resources
available to throttled utilities as defined by the util_impact_lim.
Chapter 3. Storage layout 89
For our rebalancing, we use the SET UTIL_IMPACT_PRIORITY command to set the
priority of the rebalancing process so that the impact of rebalancing is controlled.
Here are the steps:
1. Run db2 list utilities show detail to obtain the utility ID of the
rebalancing process. This command shows the current progress of the
rebalancing process in terms of estimated percentage complete.
2. Set the priority of this process by executing the SET UTIL_IMPACT_PRIRITY FOR
<utility_id> TO <priority> command. If rebalance still incurs too much
overhead or takes an excessive amount of time to complete, an offline
approach (namely backup/redirected restore) can be used. The idea is to
back up the existing database, restore the database with REDIRECT option,
add the new storage path, and continue the restore.
Similar to adding a new storage path, automatic storage allows us to remove
existing storage paths from a database or move the data off the storage paths
and rebalance them. For more details about how this can be done, see the
DB2 v9.7 Information Center at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp?topic
=/com.ibm.db2.luw.doc/welcome.html
An important consideration for storage paths or file systems used for
automatic storage is that the file systems are uniform in capacity and exhibit
similar I/O characteristics. Figure 3-3 on page 91 illustrates how containers of
a table space grow in the storage paths in which capacity is uneven.
90 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 3-3 How does a table space grow?

3. The table space starts out with two containers (/path1and /path2) that have
not yet reached maximum capacity. A new storage path (/path3) is added to
the database using the ALTER DATABASE statement. This new storage path is
not yet being used without initiating a rebalance process by ALTER
TABLESPACE.
4. The original containers in /path1 and /path2 reach the maximum capacity.

5. Because there is no storage space left in /path1, a new stripe set of

containers are added to /path2 and /path3 and they start to fill up with data.

Note: Stripe is defined as the total of all the segments for one pass of all
the back-end data disks.

6. The containers in the new stripe set (in /path2and /path3) reach their
maximum capacity.
7. A new stripe set is added only to /path3because there is no room for the
container in /path2 to grow.

Chapter 3. Storage layout 91

 As observed, we have more data in /path2after step 4. This immediately creates
a performance hot spot. Hot space can also be introduced if the I/O performance
in a storage path is significantly slower than the others. Furthermore, the
example shows that without rebalance the new storage path is not used until an
existing storage path runs out of space.

The recommended practice is to have all storage paths with the same capacity
and I/O characteristics, and rebalance after a new storage path is added.
Figure 3-4 depicts the scenario that follows this practice. With rebalance and
uniform capacity, the table space grows evenly across all the storage paths. This
ensures that parallelism remains uniform and achieves the optimal I/O
performance.

Figure 3-4 Storage grows with uniform capacity storage paths and rebalance

3.2.2 Tablespace design

As mentioned, with automatic storage we do not need to define our containers
explicitly when creating table spaces. Nevertheless, automatic storage does not
preclude us from defining other types of table spaces as well as their attributes.

First, what table spaces do you need? A simple design is to have one table space
for temporary tables, one table space for data, and one table space for all
indexes. In a partitioned environment (or DPF environment), a typical approach is
92 Best Practices for DB2 on AIX 6.1 for POWER Systems
to have separate table spaces for partitioned data and non-partitioned data. The
next step is to consider what types of table spaces you need, and their attributes.
The following sections describe suggested practices and important
considerations when designing a table space.

Pagesize
Rows of table data are organized into blocks called pages. Pages can be four
sizes: 4, 8, 16, and 32 KB. Table data pages do not contain the data for columns
defined with LONG VARCHAR, LONG VARGRAPHIC, BLOB, CLOB, DCLOB, or
XML data types, unless the LOB or XML document is inlined through the use of
INLINE LENGTH option of the column. The rows in a data page, however,
contain a descriptor of these columns.

With a different pagesize, the maximum row length can vary, as shown in
Table 3-1.

Table 3-1 Page size versus row length:

pagesize row length

4 KB 4005 bytes

8 KB 8101 bytes

16 KB 16 293 bytes

32 KB 32 677 bytes

Regardless of the page size, the maximum number of rows in a data page is 255
for tables in a REGULAR table space. This number can go higher with a LARGE
table space.

For an OLTP environment, a smaller page size is usually preferable, because it

consumes less buffer pool space with unwanted rows. 8 KB is recommended.
DW workload, on the other hand, accesses a large number of consecutive rows
at a time. Thus, a larger page size is usually better because it reduces the
number of I/O requests that are required to read a specific number of rows. The
recommended value for pagesize is 16 KB. There is, however, an exception to
this. If your row length is smaller than the page size divided by the maximum
number of rows, there is consumed space on each page. In this situation, a
smaller page size might be more appropriate.
Chapter 3. Storage layout 93
LARGE versus REGULAR
For permanent data (such as table data and indexes), we can define our table
spaces of type REGULAR or LARGE.

A REGULAR table space allows tables to have up to 255 rows per data page,
and can be managed by the database manager (DMS) or system (SMS).

A table in a LARGE table space can support more than 255 rows per data page,
which can improve space use on data pages. Indexes on the table are slightly
larger because LARGE table spaces make use of large row identifiers (RIDs),
which are 2 bytes longer than regular RIDs. In terms of performance, there is not
much difference between REGULAR and LARGE table spaces. By default, DMS
table spaces are created as LARGE table spaces. For SMS table spaces,
however, the only supported table space type is REGULAR table spaces.
DMS versus SMS and automatic storage
A table space can be managed by the DMS, the SMS, or automatic storage. For
more information about this, see the following Information Center article:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp?topic=/c
om.ibm.db2.luw.admin.dbobj.doc/doc/c0055446.html
Of the three types of table spaces, automatic storage table spaces are the
easiest to set up and maintain, and are recommended for most applications.
They are particularly beneficial in the following situations:
› You have larger tables or tables that are likely to grow quickly.
› You do not want to make regular decisions about how to manage container
growth.
› You want to store other types of related objects (for example, tables, LOBs,
indexes) in other table spaces to enhance performance.
DMS table spaces are useful in the following circumstances:
› You have larger tables or tables that are likely to grow quickly.
› You want to exercise greater control over where data is physically stored.
› You want to make adjustments to or control how storage is used (for example,
adding containers)
› You want to store other types of related objects (for example, tables, LOBs,
indexes) in other table spaces to enhance performance.

SMS table spaces are useful in the following situations:

› You have smaller tables that are not likely to grow quickly.
› You want to exercise greater control over where data is physically stored.
› You want to do little in the way of container maintenance.
› You are not required to store other types of related objects (for example,
tables, LOBs, indexes) in other table spaces. (For partitioned tables only,
indexes can be stored in table spaces separately from table data).
94 Best Practices for DB2 on AIX 6.1 for POWER Systems
In general, automatic storage table spaces and DMS table spaces have better
performance than SMS table spaces.

A DMS table space has a choice of creating its containers as a raw device or a
file in a file system. A raw device traditionally provides better performance.
Nonetheless, the concurrent I/O (CIO) feature on AIX now has almost completely
eliminated the need to use a raw device for performance. (CIO is discussed in
“File system caching” on page 97.) File systems also provides superior
manageability as compared to raw devices. Therefore, the general
recommendation is to use file systems instead of raw devices.

EXTENSIZE
The EXTENTSIZE for a table space specifies the number of pages that is written
to a container before skipping to the next container (if there are more than one
container in the table space).

To maximize I/O performance, the EXTENTSIZE is set to the number of pages

that includes an entire RAID stripe. For example, on a system with a 128 KB
segment size that uses RAID5 7+P, the stripe size is 896 KB (128 KB x 7). Note
that when calculating the RAID 5 stripe size, the Parity Disk is excluded from the
calculation.

So in this example, if the page size used is 16 KB, the extensize in pages is
896/16 = 56. The concept of segment and RAID stripe is discussed further in
3.3.3, “Storage structure: RAID levels” on page 104 and 3.3.4, “Logical drives
(LUN) and controller ownership” on page 114.

We need to take into consideration if we have many tables in which the amount of
data is far less than the EXTENTSIZE. This is not uncommon in the OLTP
environments. Consider an example where we have 40,000 tables, of which
almost 20,000 tables are empty. Because the EXTENTSIZE is the minimum
allocation for tables, a lot of space is wasted. In such environments, a smaller
EXTENSIZE might be preferable. For instance, you might consider the
EXTENTSIZE to be equal to the segment size.

PREFETCHSIZE
The PREFETCHSIZE specifies the number of pages to read by a query prior to
the pages being referenced by the query, so that the query does not need to wait
for IO. For example, suppose you have a table space with three containers. If you
set the PREFETCHSIZE to be three times the EXTENTSIZE, the database
manager can do a big-block read from each container in parallel, thereby
significantly increasing I/O throughput. This assumes that the best practice is
followed in such a way that each container is resided on a separate physical
device.
Chapter 3. Storage layout 95
The recommendation is to leave it as AUTOMATIC. This way, DB2 updates the
prefetchsize automatically when there is a change in the number of containers in
a table space. The calculation of the PREFETCHSIZE is as follows:
number_of_containers × number_of_disks_per_container × extent_size

For example, assume the extent size for a database is eight pages, and that
there are four containers, each of which exists on a single physical disk. Setting
the prefetch size to: 4 × 1 × 8 = 32 results in a prefetch size of 32 pages in total.
These 32 pages are read from each of the four containers in parallel.

The default value for number_of_disks_per_container is 1. Nonetheless, this

value can be adjusted by the DB2 registry variable DB2_PARALLEL_IO.

OVERHEAD and TRANSFERRATE

When creating a table space, we can define the OVERHEAD and
TRANSFERRATE associated the containers in the table space.
› OVERHEAD provides an estimate of the time (in milliseconds) that is required
by the container before any data is read into memory.
› TRANSFERRATE provides an estimate of the time (in milliseconds) that is
required to read one page of data into memory.

These two values affect how DB2 optimizer selects the optimal access plans for
queries. You can use the formula in Example 3-1 to estimate the values for
OVERHEAD and TRANSFERRATE.

Example 3-1 Estimate the OVERHEAD and TRANSFERRATE

OVERHEAD = average seek time in milliseconds + (0.5 * rotational latency)

where:
- 0.5 represents the average overhead of one half rotation
- rotational latency (in milliseconds) is calculated for each full rotation: (1 /
RPM) * 60 * 1000 where RPM represents rotation per minute.

For example, assume that a disk performs 7200 RPM. Using the
rotational-latency formula:
(1 / 7200) * 60 * 1000 = 8.328 milliseconds
This value can be used to estimate the overhead as follows, assuming an average
seek time of 11 milliseconds:
OVERHEAD = 11 + (0.5 * 8.328)
= 15.164
96 Best Practices for DB2 on AIX 6.1 for POWER Systems
As to TRANSFERRATE, if each tablespace container is a single physical disk, you
can use the following formula to estimate the transfer cost in milliseconds per
page:

TRANSFERRATE = (1 / spec_rate) * 1000 / 1024000 * pagesize where:

- spec_rate: represents the disk specification for the transfer rate (in megabytes
per second)

For example, suppose the specification rate for a disk is 3 megabytes per second.
Then:
TRANSFERRATE = (1 / 3) * 1000 / 1024000 * 4096
= 1.333248
or about 1.3 milliseconds per page.

If the table space containers are not single physical disks, but are arrays of disks
(such as RAID), you must take additional considerations into account when
estimating the TRANSFERRATE.

If the array is relatively small, you can multiply the spec_rate by the number of
disks, assuming that the bottleneck is at the disk level.
However, if the array is large, the bottleneck might not be at the disk level, but at
one of the other I/O subsystem components, such as disk controllers, I/O busses,
or the system bus. In this case, you cannot assume that the I/O throughput
capacity is the product of the spec_rate and the number of disks. Instead, you
must measure the actual I/O rate (in megabytes) during a sequential scan. For
example, a sequential scan resulting from select count(*) from big_table could be
several megabytes in size. In this case, divide the result by the number of
containers that make up the table space in which BIG_TABLE resides. Use this
result as a substitute for spec_rate in the formula given above.

Containers assigned to a table space can reside on other physical disks. For the
best results, all physical disks used for a given table space must have the same
OVERHEAD and TRANSFERRATE characteristics. If these characteristics are
not the same, use average values when setting OVERHEAD and
TRANSFERRATE.

File system caching

The operating system, by default, caches file data that is read from and written to
disk. For example, a typical read operation involves reading the data from a disk
into the file system cache, and then copying the data from the cache to the
application buffer (in our case, DB2 bufferpool). A similar process (but in an
opposite direction) is applied to a write operation. In a few cases, this double

Chapter 3. Storage layout 97

 buffering effect (for example, caching at both file system cache and DB2
bufferpool levels) causes performance degradation, because extra CPU cycles
need to be incurred. Because we waste memory to buffer the same page twice,
the potential paging can further negatively affect the DB2 application
performance.

AIX provides a feature to disable file system caching and avoid double caching.
This feature is known as Concurrent I/O (CIO). Another advantage of CIO is that
this feature might help reduce the memory requirements of the file system cache,
making more memory available for other uses. In DB2, we can enable CIO on
table spaces by specifying the clause NO FILE SYSTEM CACHING in the
CREATE TABLESPACE or ALTER TABLESPACE statement. (Starting from DB2
v9.5, NO FILE SYSTEM CACHING is the default for any DMS table space
created.)

LOBS (Large Objects) are not cached in DB2 bufferpool. For table spaces that
contain a fair amount of LOBS data, performance might be improved by enabling
file system caching.

3.2.3 Other considerations for DW environments

There are two more considerations for DW environments to discuss, namely
placing the tables into table spaces and compressing the data.

How to place tables into table spaces

In a DPF environment (especially for DW workloads), refer to the following
guidelines when deciding how to place tables into table spaces:
› Large tables
It is recommended that you place these tables in a separate table space
across all data partitions. These tables might benefit from the use of a
separate multi-partition table space for the indexes associated with the tables.
Range-partitioned tables might benefit from using one table space per range.
› Medium-sized tables
These tables are logically grouped together and placed into table spaces with
five or more tables in each table space. These table spaces are placed across
all data partitions. Tables that are placed in the same table space have a
logical relationship to one another (such as all coming from the same
department or requiring the same backup time frame). The indexes for these
tables can be placed within the data table space or separated into another
shared table space for indexes.
98 Best Practices for DB2 on AIX 6.1 for POWER Systems
› Small tables
Many tables are placed into one single-partition table space typically on the
catalog partition. The indexes associated with these table spaces generally
reside within the data table space.
› Replicated tables
The single-partition tables that are used for joins with multi-partition tables
can be replicated across all data partitions. These replicated tables are
placed into a separate table space.
› Materialized query tables (MQTs) and other database objects
These type of objects are typically placed in separate table spaces that are
placed across all data partitions.

Compression
Data compression provides several benefits. The most significant benefit is
lowered overall space requirements. Typical compression ratios for data tables
are between 50% and 65%, although the compression ratio can be much higher
or lower depending on the data. Another benefit of data compression is the
potential for improved performance. Processing compressed data uses more
processor cycles than uncompressed data, but requires less I/O.

Row compression, a feature available since DB2 V9.1, allows data tables to be
compressed. In DB2 V9.7, two new compression techniques are introduced:
› Index compression
Index compression compresses indexes (including indexes on declared or
created temporary tables). If a data table is compressed, new indexes created
on this table are automatically compressed. You can always enable or disable
index compression on any indexes explicitly.
› Temporary table compression.
Temporary table compression, on the other hand, compresses temporary
tables, such as created global temporary tables (CGTT) and declared global
temporary tables (DGTT). Unlike row or index compression, temporary table
compression is always on.

For more details about how row and index compression can be used, see the
DB2 Information Center at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp?topic=/c
om.ibm.db2.luw.doc/welcome.html

Important: You must apply for the IBM DB2 Storage Optimization Feature
license to use any of the compression features.
Chapter 3. Storage layout 99
3.3 Storage hardware
In the previous section, we described general considerations in DB2 storage
design. In this section, we look at storage hardware and discuss how storage
hardware can be configured and tuned to meet database storage requirements.
For illustrative purposes, we focus on the IBM System Storage™ DS5000 series.

3.3.1 Introduction to the IBM Storage DS5000 Storage Server

IBM has brought together a broad selection of storage servers into one family,
known as the DS family, to help small to large enterprises select the right
solutions for their needs. The DS family combines the high-performance IBM
System Storage DS8000 series of enterprise servers with the IBM System
Storage DS4000/5000 series of mid-range systems, and other line-of-entry
systems, namely IBM System Storage DS3000 series.

For the mid-range systems, the DS5000 series, in particular, provides an

increase in performance by using the latest technology and increased storage
capacity of the high-end DS4000® series. Our Infosphere warehouse solution
also makes use of the DS5000 series together with DB2 to provide data
warehouse solution for large enterprise.

The DS5000 series of storage servers uses Redundant Array of Independent

Disks (RAID) technology. RAID technology is used to protect the user data from
disk drive failures. DS5000 storage servers contain Fibre Channel (FC)
interfaces to connect both the host systems and disk drive enclosures. Its host
ports support existing Gbps FC SANs (storage area networks), as infrastructure
changes ports can be changed by adding, replacing, or mixing host interfaces to
support 8 Gbps FC and 10 Gbps iSCSI.

The DS5000 Storage Manager software, which comes with the hardware, can be
used to configure, manage, and troubleshoot the DS5000 storage servers. We
use this software to configure RAID arrays and logical drives, assign logical
drives to hosts, replace and rebuild failed disk drives, expand the size of the
arrays and logical drives, and convert from one RAID level to another.

In the next sections, we talk about important aspects of the DS5000 series that
deserve special attention. More importantly, a few of them have an impact on
performance of our database systems. For other details on the DS5000 series,
readers are encouraged to refer to IBM RedBooks publication Introduction to the
IBM System Storage DS5000 Series, SG24-7676.
100 Best Practices for DB2 on AIX 6.1 for POWER Systems
3.3.2 Physical components considerations
In this section we discuss the considerations for physical components.

Cables and connectors

In FC technology, frames are moved from source to destination using gigabit
transport, which is a requirement to achieve fast transfer rates. To communicate
with gigabit transport, both sides have to support this type of communication.

The FC standard specifies a procedure for speedy auto-detection. Therefore, if a

2 Gbps port on a switch or device is connected to a 1 Gbps port, it must
negotiate down and the link runs at 1 Gbps. If there are two 2 Gbps ports on
either end of a link, the negotiation runs the link at 2 Gbps if the link is up to
specifications.

A link that is too long can end up running at 1 Gbps, even with 2 Gbps ports at
either end, so be aware of your distances and make sure your fiber is good. The
same rules apply to 4 Gbps devices relative to 1 Gbps and 2 Gbps environments.
The 4 Gbps devices have the ability to negotiate back down to either 2 Gbps or 1
Gbps, depending upon the attached device and link quality.

Host Based Adapter

There are eight (DS5100 system) to sixteen (DS5300 system) host connections.
Each connection can operate at 4 Gbps, but also auto-negotiates to support 2
Gbps and 1 Gbps connections, as described previously. Each host is associated
with a Host Based Adapter (HBA). Having sixteen independent host ports allows
us to establish fully redundant direct connections to up to eight hosts. For
example,

Figure 3-5 on page 102 shows that the host ports are connected to eight sets of
FC HBA.
Chapter 3. Storage layout 101
 HBA2

HBA1

Host AIX

Figure 3-5 Directly connected hosts to the DS5000 storage server

When multiple HBAs are installed on a host, multipath driver might be used to
achieve redundancy (failover) or load sharing. We revisit this topic later in this
section.

Drives
The speed and the type of the drives2 used impacts the performance. Typically,
the faster the drive, the higher the performance. This increase in performance
comes at a cost. The faster drives typically cost more than the lower performance
drives. FC drives outperform the SATA drives. In particular, the DS5000 storage
server supports the following types of FC drives:
› 4 Gbps FC, 146.8 GB / 15K Enhanced Disk Drive Module
› 4 Gbps FC, 300 GB / 15K Enhanced Disk Drive Module
› 4 Gbps FC, 450 GB / 15K Enhanced Disk Drive Module

Note that a FC 15K drive rotates 15,000 times per minute. Also, its read and
write bandwidth are 76 Mbps and 71 Mbps, respectively.

The speed of the drive is the number or revolutions per minute (RPM). A 15 K
drive rotates 15,000 times per minute. With higher speeds, the drives tend to be
denser, as a large diameter plate driving at such speeds is likely to wobble. With
the faster speeds comes the ability to have greater throughput.

2 The words “drive” and “disk” are used interchangeably in this chapter.
102 Best Practices for DB2 on AIX 6.1 for POWER Systems
Seek time is the measure of how long it takes for the drive head to move to the
correct sectors on the drive to either read or write data. It is measured in
milliseconds (ms). The faster the seek time, the quicker data can be read from or
written to the drive. The average seek time reduces when the speed of the drive
increases. Typically, a 7.2 K drive has an average seek time of around 9 ms, a 10
K drive has an average seek time of around 5.5 ms, and a 15 K drive has an
average seek time of around 3.5 ms. Together with RPM, the seek time is used to
determined the value of OVERHEAD for our DB2 table spaces (see 3.2.2,
“Tablespace design” on page 92).

Command queuing allows for multiple commands to be outstanding to the disk

drive at the same time. The drives have a queue where outstanding commands
can be dynamically rescheduled or re-ordered, along with the necessary tracking
mechanisms for outstanding and completed portions of workload. The FC disks
currently have a command queue depth of 16.

Hot spare drive

A hot spare drive is like a replacement drive installed in advance. Hot spare disk
drives provide additional protection that might prove to be essential in case of a
disk drive failure in a fault tolerant array.

There is no definitive recommendation as to how many hot spares you must

install, but it is common practice to use a ratio of one hot spare for 28 drives.

We recommend that you also split the hot spares so that they are not on the
same drive loops (see Figure 3-6 on page 104).

Tip: When assigning disks as hot spares, make sure they have enough
storage capacity. If the failed disk drive is larger than the hot spare,
reconstruction is not possible. Ensure that you have at least one of each size
or all larger drives configured as hot spares.
Chapter 3. Storage layout 103
Figure 3-6 Hot spare coverage with alternating loops

3.3.3 Storage structure: RAID levels

An array is a set of drives that the system logically groups together to provide one
or more logical drives to an application host or cluster. RAID is the technology
that we use to form an array. When defining arrays, you often have to
compromise among capacity, performance, and redundancy.

I/O characteristics between OLTP and DW workloads

To better understand what it is meant by the performance impact of other RAID
levels, we first need to understand the I/O characteristics that OLTP and DW
workloads or applications exhibit.

OLTP (Online Transaction Processing)

OLTP workloads, often described as transaction-based environments, are
typically I/O intensive. They require a high number of I/O per seconds (IOPS).
Their I/O operations are geared towards random reads and writes, and use a
small random data block pattern to transfer data. The latter benefits by having
more back-end drives, because a large number of back-end drives enable more
host I/Os to be processed simultaneously, as read cache is far less effective, and
the misses need to be retrieved from disk.
104 Best Practices for DB2 on AIX 6.1 for POWER Systems
DW (Data Warehouse)
DW workloads typically require massive amounts of data sent and generally use
large sequential blocks to reduce disk latency. This environment is also known as
a throughput-based environment. Read operations make use of the cache to
stage greater chunks of data at a time, to improve the overall performance.
Throughput rates are heavily dependent on the storage server’s internal
bandwidth.

RAID levels
In this section, we discuss the RAID levels and explain why we choose a
particular setting in a particular situation. You can draw your own conclusions.

RAID 0: For performance, but generally not recommended

RAID 0 (Figure 3-7) is also known as data striping. It is well-suited for program
libraries requiring rapid loading of large tables, or more generally, applications
requiring fast access to read-only data or fast writing. RAID 0 is designed to
increase performance. There is no redundancy, so any disk failures require
reloading from backups. Select RAID 0 for applications that benefit from the
increased performance capabilities of this RAID level. Never use this level for
critical applications that require high availability.

Logical Drive

Host view

Disk 1 Disk 2 Disk 3

Actual
device
mappings

Stripeset

Figure 3-7 RAID 0

Chapter 3. Storage layout 105
RAID 1: For availability/good read response time
RAID 1 (Figure 3-8) is also known as disk mirroring. It is most suited to
applications that require high data availability, good read response times, and
where cost is a secondary issue. The response time for writes can be somewhat
slower than for a single disk, depending on the write policy. The writes can either
be executed in parallel for speed or serially for safety. Select RAID 1 for
applications with a high percentage of read operations and where the cost is not
the major concern.

Logical Drive
Block 0
Block 1
Host view
Block 2
etc.

Actual
device Block 0 Block 0
mappings Block 1 Block 1
Block 2 Block 2
etc. Mirrorset etc.

Figure 3-8 RAID 1

Because the data is mirrored, the capacity of the logical drive when assigned
RAID 1 is 50% of the array capacity.

The following list details recommendations when using RAID 1:

› Use RAID 1 for the disks that contain your operating system. It is a good
choice, because the operating system can usually fit on one disk.
› Use RAID 1 for transaction logs. Typically, the database server transaction log
can fit on one disk drive. In addition, the transaction log performs mostly
sequential writes. Only rollback operations cause reads from the transaction
logs. Therefore, we can achieve a high rate of performance by isolating the
transaction log on its own RAID 1 array.
106 Best Practices for DB2 on AIX 6.1 for POWER Systems
› Use write caching on RAID 1 arrays. Because a RAID 1 write does not
complete until both writes have been done (two disks), performance of writes
can be improved through the use of a write cache. When using a write cache,
be sure it is battery-backed up.

Note: RAID 1 is actually implemented only as RAID 10 (see “RAID 10: Higher
performance than RAID 1” on page 109) on the DS5000 storage server
products.

RAID 5: High availability and fewer writes than reads

RAID 5 (Figure 3-9) stripes data and parity across all drives in the array. RAID 5
offers both data protection and increased throughput. When you assign RAID 5
to an array, the capacity of the array is reduced by the capacity of one drive (for
data-parity storage). RAID 5 gives you higher capacity than RAID 1, but RAID
level 1 offers better performance.

Logical Drive
Block 0
Block 1
Block 2
Block 3 Host View
Block 4
Block 5
etc.

Block 0 Block 1 Block 2 Block 3 Parity 0-3

Block 5 Block 6 Block 7 Parity 4-7 Block 4
Block 10 Block 11 Parity 8-11 Block 8 Block 9
Block 15 Parity 12-15 Block 12 Block 13 Block 14

Figure 3-9 RAID 5

RAID 5 is best used in environments requiring high availability and fewer writes
than reads.

RAID 5 is good for multi-user environments, such as database or file system

storage, where typical I/O size is small, and there is a high proportion of read
activity. Applications with a low read percentage (write-intensive) do not perform
Chapter 3. Storage layout 107
as well on RAID 5 logical drives because of the way a controller writes data and
redundancy data to the drives in a RAID 5 array. If there is a low percentage of
read activity relative to write activity, consider changing the RAID level of an array
for faster performance.

Use write caching on RAID 5 arrays, because RAID 5 writes are not completed
until at least two reads and two writes have occurred. The response time of
writes is improved through the use of write cache (be sure it is battery-backed
up). RAID 5 arrays with caching can give as good as performance as any other
RAID level, and with a few workloads, the striping effect gives better performance
than RAID 1.

RAID 6: High availability with additional fault tolerance

RAID 6 (Figure 3-10) is a RAID level employing n+2 drives, which can survive the
failure of any two drives. RAID 6 stripes blocks of data and parity across an array
of drives. It calculates two sets of information for each block of data (p+q). For the
purposes of RAID 6 p+q, they can be used to generate up to two missing values
from a set of data. The key to this method is the q, which is a codeword based
upon Reed-Solomon error correction. As such, q is more like a CRC than parity.
The calculation of q is complex. In the case of the DS5000 storage server, this
calculation is made by the hardware.

Figure 3-10 RAID 6

108 Best Practices for DB2 on AIX 6.1 for POWER Systems
By storing two sets of distributed parities, RAID 6 is designed to tolerate two
simultaneous disk failures. Furthermore, unlike RAID 5, RAID 6 is also protected
from data loss during array re-construction.

Due to the added impact of more parity calculations, RAID 6 is slightly slower
than RAID 5 in terms of writing data. Nevertheless, there is essentially no impact
on read performance when comparing between RAID 5 and RAID 6 (provided
that the number of disks in the array is equal).

RAID 10: Higher performance than RAID 1

RAID 10 (refer to Figure 3-11), also known as RAID 1+0, implements block
interleave data striping and mirroring. In RAID 10, data is striped across multiple
disk drives. Those drives are then mirrored to another set of drives.

Logical Drive
Block 0
Block 1
Block 2
Block 3
Block 4
Block 5
etc.

Controller
internal
mapping
Actual
device
mappings

Disk 1 Disk 2 Disk 3 Disk 4 Disk 5 Disk 6

Block 0 Block 0 Block 1 Block 1 Block 2 Block 2
Block 3 Block 3 Block 4 Block 4 Block 5 Block 5
Block 6 Block 6 Block 7 Block 7 Block 8 Block 8

Stripeset

Figure 3-11 RAID 10

The performance of RAID 10 is approximately the same as RAID 0 for sequential

I/Os. RAID 10 provides an enhanced feature for disk mirroring that stripes data
and copies the data across all the drives of the array. The first stripe is the data
stripe. The second stripe is the mirror (copy) of the first data stripe, but it is
shifted over one drive. Because the data is mirrored, the capacity of the logical
drive is 50% of the physical capacity of the hard disk drives in the array.
Chapter 3. Storage layout 109
 The recommendations for using RAID 10 are as follows:
› Use RAID 10 whenever the array experiences more than 10% writes. RAID 5
does not perform as well as RAID 10 with a large number of writes.
› Use RAID 10 when performance is critical. Use write caching on RAID 10.
Because a RAID 10 write is not completed until both writes have been done,
write performance can be improved through the use of a write cache (be sure
it is battery-backed up).

When comparing RAID 10 to RAID 5:

› RAID 10 writes a single block through two writes. RAID 5 requires two reads
(read original data and parity) and two writes. Random writes are significantly
faster on RAID 10.
› RAID 10 rebuilds take less time than RAID 5 rebuilds. If a real disk fails, RAID
10 rebuilds it by copying all the data on the mirrored disk to a spare. RAID 5
rebuilds a failed disk by merging the contents of the surviving disks in an array
and writing the result to a spare.
› RAID 10 is the best fault-tolerant solution in terms of protection and
performance, but it comes at a cost. You must purchase twice the number of
disks that are necessary with RAID 0.

A comparison of RAID levels

Table 3-2 compares the RAID levels.

Table 3-2 RAID levels comparison

RAID Description Workload Advantage Disadvantage

0 Stripes data across OLTP Performance, due to parallel No redundancy. If one

multiple drives. DW operation of the access. drive fails, the data is
lost.

1 The disk's data is OLTP Performance, as multiple Storage costs are

mirrored to another requests can be fulfilled doubled.
drive. simultaneously.

10 Data is striped across OLTP Performance, as multiple Storage costs are

multiple drives and requests can be fulfilled doubled.
mirrored to the same simultaneously. Most reliable
number of disks. RAID level on the DS5000
storage server.

5 Drives operate OLTP Good for reads, small IOPS, Writes are particularly
independently with DW many concurrent IOPS, and demanding.
data and parity blocks random I/Os.
distributed across all
drives in the group.
110 Best Practices for DB2 on AIX 6.1 for POWER Systems
RAID Description Workload Advantage Disadvantage

6 Stripes blocks of data OLTP Good for multi-user Slower in writing data,
and parity across an DW environments, such as complex RAID controller
array of drives and database, where typical I/O architecture.
calculates two sets of size is small, and in situations
parity information for where additional fault
each block of data. tolerance is required.

Array configuration
Before you can start using the physical disk space, you must configure it. Based
on the previous recommendation in RAID levels, you divide your (physical) drives
into arrays accordingly and create one or more logical drives inside each array.

In simple configurations, you can use all of your drive capacity with one array and
create all of your logical drives in that unique array. However, this presents the
following drawbacks:
› If you experience a (physical) drive failure, the rebuild process affects all
logical drives, and the overall system performance goes down.
› Read/write operations to logical drives are still being made to the same set of
physical hard drives.

The array configuration is crucial to performance. You must take into account all
the logical drives inside the array, as all the logical drives inside the array impact
the same physical disks. If you have two logical drives inside an array and they
both are high throughput, there might be contention for access to the physical
drives as large read or write requests are serviced. It is crucial to know the type
of data that each logical drive is used for and try to balance the load so
contention for the physical drives is minimized. Contention is impossible to
eliminate unless the array only contains one logical drive.

Number of drives
The more physical drives you have per array, the shorter the access time for read
and write I/O operations.

You can determine how many physical drives is associated with a RAID controller
by looking at disk transfer rates (rather than at the megabytes per second). For
example, if a disk drive is capable of 75 nonsequential (random) I/Os per second,
about 26 disk drives working together can, theoretically, produce 2000
nonsequential I/Os per second, or enough to hit the maximum I/O handling
capacity of a single RAID controller. If the disk drive can sustain 150 sequential
I/Os per second, it takes only about 13 disk drives working together to produce
the same 2000 sequential I/Os per second and keep the RAID controller running
at maximum throughput.
Chapter 3. Storage layout 111
 Tip: Having more physical drives for the same overall capacity gives you:
› Performance
By doubling the number of the physical drives, you can expect up to a 50%
increase in throughput performance.
› Flexibility
Using more physical drives gives you more flexibility to build arrays and
logical drives according to your needs.
› Data capacity
When using RAID 5 logical drives, more data space is available with
smaller physical drives because less space (capacity of a drive) is used for
parity.

Enclosure loss protection

Enclosure loss protection is a good way to make your system more resilient
against hardware failures. Enclosure loss protection means that you spread your
protection arrays across multiple enclosures rather than in one enclosure so that
a failure of a single enclosure does not take a whole array offline.

Figure 3-12 on page 113 shows an example of the enclosure loss protection. If
enclosure number 2 were to fail, the array with the enclosure loss protection still
functions (in a degraded state), as the other drives are not affected by the failure.
112 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 3-12 Enclosure loss protection

DB2 database considerations

Further considerations on configuring an array are needed for a DB2 database.

OLTP workloads
OLTP environments contain a fairly high level of reads and a considerable
amount of writes. In most cases, it has been found that laying out the tables
across a number of logical drives that were created across several RAID 5 arrays
of 8+1 parity disks, and configured with a segment size of 64 KB or 128 KB, is a
good starting point to begin testing. This configuration, coupled with host
recommendations to help avoid offset and striping conflicts, seems to provide a
good performance start point to build from. Remember that high write
percentages might result in a need to use RAID 10 arrays rather than the RAID 5.
This is environment-specific and requires testing to determine. A rule of thumb is
that if there are greater than 25–30% writes, then you might want to look at RAID
10 over RAID 5.

Suggested practice: Spread the containers across as many drives as

possible, and ensure that the logical drive spread is evenly shared across the
DS5000 storage server’s resources. Use multiple arrays where larger
containers are needed.
Chapter 3. Storage layout 113
 DW workloads
In DW environments, write activity is relatively low, as are the random I/Os. One
recommendation is that for each data partition, we allocate eight drives (146 GB
FC drives) and group them into one 7 + 1 RAID 5 array.

Logs and archives

The DB2 logs and archive files generally are high write workloads, and
sequential in nature. We recommend that they be placed on RAID 10 logical
drives.

As these are critical files to protect in case of failures, we recommend that you
keep two full copies of them on separate disk arrays in the storage server. This is
to protect you from the unlikely occurrence of a double disk failure, which can
result in data loss. Also, as these are generally smaller files and require less
space, we suggest that two separate arrays of 1+1 or 2+2 RAID 1 be used to
hold the logs and the mirror pair separately.

3.3.4 Logical drives (LUN) and controller ownership

A logical drive, sometimes referred to as LUNs (LUN stands for Logical Unit
Number and represents the number a host uses to access the logical drive), are
defined over an array and have a defined RAID level and capacity. More
specifically, a logical driver stripes across all data disks in the array, which can be
equal to the entire array, or a portion of the array. The latter means that multiple
logical drives are allowed to reside on the same array. However, they have to
share I/Os bandwidth on the same set of disks. Our recommendation is to define
one logical drive (or LUN) over an array.

Each logical drive has a preferred controller of ownership. This controller

normally handles all I/O requests for this particular logical drive. In other words,
each logical drive is owned by only one controller. The alternate controller takes
over and handles the I/O requests in case of a failure along the I/O path. When
defining logical drives, the system normally alternates ownership between the
two controllers as they are defined.

As a suggested practice, configuring logical drives to spread evenly among the

two controllers (in a DS5000 storage server) allows better load balancing and
controller use.

The segment size is the maximum amount of data that is written or read from a
disk per operation before the next disk in the array is used. For OLTP
environments, we suggest that the segment size be 64 KB to 128 KB.
114 Best Practices for DB2 on AIX 6.1 for POWER Systems
 With DW workloads, the focus is on moving high throughput in fewer I/O, and the
nature of these workloads is generally sequential in nature. As a result, you want
to have larger segments (128 KB or higher) instead to get the most from each
stripe. When creating a logical drive, we specify the stripe width to be the amount
of segments size multiplied by the number of disks in the logical drive (or array).
More importantly, stripe width is used to determine our EXTENTSIZE during the
creation of our table spaces in DB2 (refer to 3.2.2, “Tablespace design” on
page 92 for more details).

3.4 Tuning storage on AIX

Before starting the creation of volume groups or logical volumes, configure and
tune the storage at the AIX level.

3.4.1 Multipath driver

AIX offers a multipath driver, called Multiple Path I/O (MPIO). It allows I/O load
balancing and automatic path failover. With MPIO, a device can be uniquely
detected through one or more physical connections, or paths. A path-control
module (PCM) provides the path management functions.

An MPIO-capable device driver can control more than one type of target device.
A PCM can support one or more specific devices. Therefore, one device driver
can be interfaced to multiple PCMs that control the I/O across the paths to each
of the target devices.

The AIX PCM has a health-check capability that can be used to do the following
tasks:
› Check the paths and determine which paths are currently usable for sending
I/O.
› Enable a path that was previously marked failed because of a temporary path
fault (for example, when a cable to a device was removed and then
reconnected).
› Check currently unused paths that are used if a failover occurred (for
example, when the algorithm attribute value is failover, the health check can
test the alternate paths).

MPIO is part of the AIX operating system and does not need to be installed
separately. The required AIX 6.1 level is TL0 (IZ13627).
Chapter 3. Storage layout 115
3.4.2 hdisk tuning
On AIX, each logical drive or LUN is called a physical volume (PV) and often
referred to as hdiskx (where x is a unique integer on the system).

There are three parameters that are important to set for hdisks for performance:
› queue_depth
› max_transfer
› max_coalesce

queue_depth
The maximum queue depth or queue_depth is the maximum number of
concurrent operations in progress on the physical device. Excess requests
beyond queue_depth are held on a pending queue in the disk driver until an
earlier request completes. Setting the queue_depth for every hdisk in the system
to the appropriate value is important for system performance. The valid values for
queue_depth are from 1 to 256.

Typically, disk vendors supply a default queue_depth value appropriate for their
disks in their disk-specific ODM device support fileset. If you know this number
you can use the exact number for the queue_depth calculation. If not, you can
use a number between four and 16, which is the standard queue depths for
drives. FC drives usually have a queue depth of 16. You also need to know how
many disks you have per hdisk. For example, if you have a RAID 5 7+P
configuration, you have eight disks per hdisk so you can start with a
queue_depth for the hdisk of 16 * 8 = 128. You can monitor the queues using
iostat -D. OLTP environments usually require higher queue_depth to perform
better than DW environments. To set a queue_depth of 64, issue the following
command:
#chdev -l hdiskX -a queue_depth=64 -P

To make the attribute changes permanent, use the -P option to update the AIX
ODM attribute. To make this change there must either be no volume group
defined on the hdisk, or the volume group that the hdisk belongs to must be
varied off. You need to reboot the server to make this change take effect.
116 Best Practices for DB2 on AIX 6.1 for POWER Systems
When setting this value, keep the following in mind:
› Setting queue_depth too high might result in the storage device being
overwhelmed. This can result in a range of issues from higher I/O latency
(lower performance) or I/Os being rejected or dropped by the device (errors in
the AIX errpt and greatly reduced performance).
› Setting queue_depth too high for FC disks might overrun the maximum
number of commands (num_cmd_elems) allowed outstanding by the FC driver
(see 3.4.3, “Fibre Channel adapters configuration” on page 118 on how to
tune num_cmd_elems).
› A combination of a large number of large I/Os for FC disks might overrun the
PCI bus address space available for the FC port.
› queue_depth is not honored for disks that do not support SCSI Command
Tagged Queueing (CTQ). Also, queue_depth is not honored for disks whose
q_type attribute is set to none or whose SCSI INQUIRY responses indicate
lack of support for CTQ.

max_transfer
The maximum transfer size (max_transfer) sets the limit for the maximum size of
an individual I/O to the disk. Although applications might make larger I/O
requests, those requests are broken down into multiple max_transfer-sized
requests before they are handed to the disk driver.

For OLTP workloads, most I/Os are for discontiguous 4 or 8 K pages (depending
on the database page size). If it is a DW workload, then most likely there are lots
of scans and I/O sizes are larger. In this case a larger max_transfer means fewer
round trips if the I/Os are large and contiguous. We recommend leaving the
default max_transfer of 256 KB (0x40000) for OLTP workloads and set it to 1 MB
(0x100000) for DW workloads. To set a max_transfer of 1 Mb issue the following
command:

#chdev -l hdiskX -a max_transfer=0x100000 -P

When setting this value, keep the following in mind:

› A combination of high queue_depth and large I/Os for FC disks might overrun
the PCI bus address space available for the FC port.
› The effective maximum transfer size for a disk is the minimum of the disk's
max_transfer tunable and the actual maximum transfer size of the underlying
adapter for all of the paths. That is, a disk cannot have a larger maximum
transfer size than the underlying adapters (see max_xfer_size in 3.4.3, “Fibre
Channel adapters configuration” on page 118).
Chapter 3. Storage layout 117
 max_coalesce
The max_coalesce value sets the limit for the maximum size of an individual I/O
that the disk driver creates by coalescing smaller, adjacent requests. We
recommend setting this value equal to max_transfer when possible. To set a
max_coalesce of 1 Mb issue the following command:

#chdev -l hdiskX -a max_coalesce=0x100000 -P

3.4.3 Fibre Channel adapters configuration

There are two parameters that are important to tune for Host Based Adapters
(HBA) for performance:
› num_cmd_elems
› max_xfer_size

num_cmd_elems
The num_cmd_elems value sets a limit for the maximum number of SCSI I/O
requests that can be active for the FC port at one time. The actual maximum
number of commands that can be issued at one time is the minimum of
num_cmd_elems and the aggregate queue_depth of the devices using the port.
Supported values are between 20 and 2048.

Any excess requests beyond num_cmd_elems are held on a pending queue in

the FC protocol driver until an earlier request completes. The fcstat No
Command Resource Count statistic tracks the number of times that a request
cannot be issued due to lack of num_cmd_elems.

We recommend setting this value to its maximum of 2048 and tuning down if
necessary. To do so, use the chdev command. For example to set this value to
2048 issue:

#chdev -l fcs0 -a num_cmd_elems=2048 -P

When setting this value keep the following in mind that a combination of a large
number of large I/Os for Fibre Channel disks might overrun the PCI bus address
space available for the Fibre Channel port.

max_xfer_size
The max_xfer_size value sets the limit for the maximum size of an individual I/O
sent by the port. This value also influences the amount of PCI bus address space
allocated for DMA mapping of I/O buffers. The default setting of 0x100000 (1 MB)
causes the FC driver to request the normal amount of PCI bus address space.
Any larger setting causes the FC driver to request a larger amount of PCI bus
118 Best Practices for DB2 on AIX 6.1 for POWER Systems
 address space. There are only two PCI bus address space sizes that the FC
driver requests. It does not request more and more as you increase
max_xfer_size further.

Any excess requests beyond the limit of the drivers ability to map for DMA is held
on a pending queue in the adapter driver until earlier requests complete. Each
time a request is set aside because it cannot be mapped for DMA a PCI_DMA
error is logged. The fcstat No DMA Resource Count statistic tracks the number of
times that a request cannot be issued due to lack of PCI address space for DMA.

We suggest leaving the default max_xfer_size for both OLTP and DW. As an
example, if you were to set this value to 2 Mb issue:
#chdev -l fcs0 -a max_xfer_size=0x200000 -P

When setting this value, keep the following issues in mind:

› The effective maximum transfer size for the port is the minimum of the port's
max_xfer_size and the overlying device's maximum transfer size. The FC
driver does not coalesce smaller requests into larger ones.
› Some PCI slots on a few machines have an overall limit on the amount of PCI
bus address space available for DMA mapping. On those machines it might
not be possible to configure both ports of a dual-port FC adapter with
increased max_xfer_size. If that happens, the Fibre Channel adapter driver
logs an error and one or both of the adapter ports remain in the Defined state.

3.5 LVM configuration: Volume groups and logical

volumes
The Logical Volume Manager (LVM) is a set of operating system commands,
library subroutines, and other tools that allow the user to establish and control
logical volume storage. The LVM controls disk resources by mapping data
between a more simple and flexible logical view of storage space and the actual
physical disks. The LVM does this using a layer of device-driver code that runs
above traditional disk device drivers. This logical view of the disk storage is
provided to the applications, and is independent of the underlying physical disk
structure.

To manage the disk storage the LVM uses a hierarchy of structures that have a
clearly defined relationship between them. The lowest element in this structure is
the PV. We have seen in previous sections how we started building the storage
layout (in Figure 3-1 on page 86 and Figure 3-2 on page 87) by taking the disks,
and creating arrays and LUNs. Each LUN is seen by the LVM as a physical
volume and has a name, usually /dev/hdiskx (where x is a unique integer on the
system).
Chapter 3. Storage layout 119
 Every physical volume in use belongs to a volume group (VG) unless it is being
used as a raw storage or a readily available spare (also known as Hot Spare). A
VG is a collection of one or more PVs. Within each volume group, one or more
logical volumes (LVs) are defined. LVs are the way to group information located
on one or more PVs. LVs are an area of disk used to store data that appears to
be contiguous to the application, but can be non-contiguous on the actual PV. It
is this definition of a LV that allows them to be extended, relocated, span
multiples PVs, and have their contents replicated.

Now that we have a basic understanding of the LVM, we look at creation and
configuration in detail.

3.5.1 Creating and configuring VGs

As discussed previously, a VG is a collection of PVs. When you install a new
system, one VG (the root VG, called rootvg) is created. New VGs can be created
with the mkvg command. It is important to remember that the rootvg has attributes
that differs from the user defined volume groups. Particularly, it cannot be
imported. We recommend you organize PVs into VGs separate from rootvg
because of the following reasons:
› For safer and easier maintenance.
– Operating system updates, reinstallations, and crash recoveries are safer
because you can separate user file systems from the operating system so
that user files are not jeopardized during these operations.
– Maintenance is easier because you can update or reinstall the operating
system without having to restore user data. For example, before updating,
you can remove a user-defined VG from the system by un-mounting its file
systems and deactivating it (using varyoffvg). If necessary, the volume
can be exported (using exportvg). After updating the system software, you
can reintroduce the user-defined VG (using importvg), activate it
(varyonvg), and then remount its file systems.
› For physical partition sizes. All PVs within the same VG must have the same
physical partition size. To have PVs with other physical partition sizes, we
need to place each size in a separate VG. We talk about what the
recommended physical partition size is for the VGs related to DB2 in “Physical
partition size” on page 121.
› For high availability (HA) purposes. Depending on the HA solution chosen, we
need to have the VG created in such a way that more than one system can
access those VGs. More on this topic in “Other considerations” on page 124.
120 Best Practices for DB2 on AIX 6.1 for POWER Systems
We want to create as few VGs as possible to make managing the resources
easier, but enough to isolate its data if needed. For a typical OLTP environment
we suggest creating one VG for data (permanent, temporal and backup), one VG
for the active logs, and one VG for the archival logs. For a DW environment we
suggest one VG for data (permanent, temporal, and backup) and active logs, and
one VG for archival logs.

The reason for having one fewer VG in DW environments is that we want the
active logs distributed in the database partitions. Because active logs are not as
performance critical as in an OLTP environment, they do not require dedicated
disks for them.

Next, we discuss performance decisions that we need to make before the actual
creation.

Physical partition size

When you add a PV to a VG, the PV is partitioned into contiguous, equal size
units of space called physical partitions (PPs). A PP is the smallest unit of
storage allocated. PVs inherit the VGs physical partition size, which can only be
set when the VG is created.

The LVM limits the number of physical partitions that a PV can have. Those limits
depend on the type of VG chosen:
› Original
› Big
› Scalable

For Original or Big VGs, the maximum number of PP per PV is 1016. That limit
can be changed by a factor to make it larger. The factor is between 1 and 16 for
Original VGs and 1 and 64 for Big VGs. The maximum number of PPs per PV for
a VG changes to 1016 * factor. When using this factor the maximum number of
PVs for that VG is reduced by MaxPVs / factor. For example, an Original VG can
have up to 32 PVs. If we decide to increase the number of PPs by a factor of 4,
we are able to have 1016 * 4 = 4064 PPs per PV for that VG, but now we only can
have 32 / 4 = 8 PVs in that VG. For Big VGs it is the same, except that the largest
number of PVs we can handle is 128, so in this example we end up with 32 PVs
as the maximum.

Scalable VGs can accommodate up to 1024 Pvs, 256 LVs and 32,768 PPs by
default. The number of LVs and PPs can be beyond the default values. In
particular, PPs can be increased up to 2,097,152 (Maximum PPs per VG is
expressed in units of 1024 PPs, the maximum being 2048, that is 1024 * 2048 =
2,097,152 PPs).
Chapter 3. Storage layout 121
The disadvantage is than increasing these defaults can significantly increase the
size of the VG Description Area (VGDA)3.

These values must only be increased as needed because they cannot be

decreased. Meanwhile, as the VGDA space increases all VGDA update
operations (creating a LV, changing a LV, adding a PV, and so on) can take longer
and longer to run.

Another consideration for the choice of VG type is that the LVM with Original and
Big VGs reserves by default the first 512 bytes of the volume for the LV control
block. Therefore, the first data block starts at an offset of 512 bytes into the
volume. Care is taken when laying out the segment size of the logical drive to
enable the best alignment. You can eliminate the LV control block on the LV by
using a Scalable VG, or by using the -T 0 option for Big VGs.

Our suggestions is to use Scalable VGs. This allows accommodating large PV

sizes, large number of PVs in the VG and allowing for future growth without the
risk of hitting the limitations that Original or Big VGs have. Furthermore, we
eliminate the risk of alignment problems as explained before.

Now that we have decided to use Scalable VGs, we need to choose the PP size
and the maximum number of PPs for the VG. The first step is to know how much
physical storage we need to accommodate under that VG. The result of PP Size
* Maximum Number of PP per VG can accommodate at least that capacity. A
best practice for Original or Big VGs for PP size is to go as small as possible
given the maximum PPs per PV restrictions. The reason for that is that PP size
has a performance impact when doing mirroring at the LVM level or in certain
cases when doing LV stripping. Because we do not recommend the usage of
mirroring at the LVM level or LV stripping of differently-sized LVs, then the choice
of a PP size only influences the minimum size allocation on the PV.

This parameter does not influence the IO performance, so it is not performance

critical to go small. So the only restriction that the PP size has in our
configuration is that we are not able to create a file system smaller than a PP.
Also, by going small on the PP size we might need to increase the maximum
number of PP per VG and that negatively impacts the performance of VGDA
operations, as explained before, so we do not want to choose a large number in
that area.

A good size for PPs is 64 Mb. Then, depending on the capacity of all the storage
in the volume group, you can calculate what the maximum number of PP per VG
is. For example, if we want to use 64 Mb PPs and a maximum number of PP per

3
The VGDA contains information about the VG, the LVs that reside in the VG, and the PVs that make
up the volume group
122 Best Practices for DB2 on AIX 6.1 for POWER Systems
VG of 512 (that is 512 * 1024 = 524,288 PPs), we can accommodate up to 32 TB
of capacity in that VG. You can use the following command to create such a VG:
mkvg -S -y <vgName> -s 64 -P 1024 f <hdiskList>

Logical track group (LTG) size

When the LVM receives a request for an I/O, it breaks the I/O down into what is
called logical track group (LTG) sizes before it passes the request down to the
device driver of the underlying disks. The LTG is the maximum transfer size of a
logical volume and is common to all the logical volumes in the volume group.

value is set by the varyonvg command using the -M flag. When the LTG size is set
using the -M flag, the varyonvg and extendvg commands might fail if an
underlying disk has a maximum transfer size that is smaller than the LTG size. To
obtain the maximum supported LTG size of your hard disk, you can use the
lquerypv command with the -M flag. The output gives the LTG size in KB, as
shown in the following example.
# /usr/sbin/lquerypv -M hdisk0

The lspv command displays the same value as MAX REQUEST, as shown in
Example 3-2.

Example 3-2 lspv command

# lspv hdisk0
PHYSICAL VOLUME: hdisk0 VOLUME GROUP: rootvg
PV IDENTIFIER: 000bc6fdbff92812 VG IDENTIFIER
000bc6fd00004c00000000fda469279d
PV STATE: active
STALE PARTITIONS: 0 ALLOCATABLE: yes
PP SIZE: 16 megabyte(s) LOGICAL VOLUMES: 9
TOTAL PPs: 542 (8672 megabytes) VG DESCRIPTORS: 2
FREE PPs: 431 (6896 megabytes) HOT SPARE: no
USED PPs: 111 (1776 megabytes) MAX REQUEST: 256 kilobytes
FREE DISTRIBUTION: 108..76..30..108..109
USED DISTRIBUTION: 01..32..78..00..00
Chapter 3. Storage layout 123
You can list the value of the LTG in use with the lsvg command, as shown in
Example 3-3.

Example 3-3 lsvg command

# lsvg rootvg
VOLUME GROUP: rootvg VG IDENTIFIER:

000bc6fd00004c00000000fda469279d
VG STATE: active PP SIZE: 16 megabyte(s)
VG PERMISSION: read/write TOTAL PPs: 542 (8672 megabytes)
MAX Lvs: 256 FREE PPs: 431 (6896 Megabytes)
LVs: 9 USED PPs: 111 (1776 megabytes)
OPEN LVs: 8 QUORUM: 2
TOTAL PVs: 1 VG DESCRIPTORS: 2
STALE PVs: 0 STALE PPs: 0
ACTIVE PVs: 1 AUTO ON: yes
MAX PPs per VG: 32512
MAX PPs per PV: 1016 MAX PVs: 32
LTG size (Dynamic): 256 kilobyte(s) AUTO SYNC: no

HOT SPARE: no BB POLICY: relocatable

Note that the LTG size for a VG is displayed as dynamic in the lsvg command
output.

Other considerations
Depending on the High Availability solution being considered, other important
considerations when creating volume groups are taken into account. Of
particular importance is the Concurrent Volume Group flag. The concurrent
access VG is a VG that can be accessed from more than one host system
simultaneously. This option allows more than one system to access the PVs.
Then, through a concurrent capable VG, they can now concurrently access the
information stored on them. Initially, this was designed for high-availability
systems, with the high-availability software controlling the sharing of the
configuration data between the systems. However, the application must control
the concurrent access to the data. Another consideration is the major number for
a VG. The major number is a numerical identifier for the VG. It is recommended
in multi-host environments that the major numbers are consistent across all
hosts.
124 Best Practices for DB2 on AIX 6.1 for POWER Systems
3.5.2 Creating and configuring logical volumes
An LV is a portion of a PV viewed by the system as a single unit. LVs consist of
LPs, each of which maps to one or more PPs. The LV presents a simple
contiguous view of data storage to the application while hiding the more complex
and possibly non-contiguous physical placement of data.

LVs can only exist within one VG. They cannot be mirrored or expanded onto
other VGs. The LV information is kept in the VGDA and the VGDA only tracks
information pertaining to one VG.

The creation of LVs allows many levels of control by the user:

› No control
› Physical volume specific control
› Partition location control

The Logical Volume Manager subsystem provides flexible access and control for
complex physical storage systems. LVs can be mirrored and stripped with other
strip sizes. There are other types of LVs:
› boot: Contains the initial information required to start the system.
› jfs: Journaled File System
› jfslog: Journaled File Systems log
› jfs2: Enhanced Journaled File System
› jfs2log: Enhanced Journaled File System log
› paging: Used by the virtual memory manager to swap out pages of memory

Following Figure 3-1 on page 86 and Figure 3-2 on page 87 we are creating
each LV associated with one and only one hdisk. Also, as we discuss in the next
section, we are creating a jfs2 file system on each of those LVs, so the jfs2 type is
specified when creating the LVs. When specifying the size of the LV, remember
that a LV is made of logical partitions and each of those logical partitions
corresponds in size to a PP on the PV (in this case only one because we do not
have mirroring). Thus, when specifying the LV size, do so in multiples of PP size
to void wasted space. The size can be specified in logical partition units (the
default) or KB/MB/GB.

The command to use to create each LV is as follows:

mklv -t jfs2 -y <lvName> <vgName> <lvSize> <hdiskName>

For example, to create one 512 GB LV on vg1 on hdisk1 issue:

mklv -t jfs2 -y lv1 vg1 512GB hdisk1

Chapter 3. Storage layout 125
 Tip: It is a best practices to create a LV, because you can give it a specific
name that is relevant to what it contains. If you do not create a LV, AIX creates
it with a generic name, if it does not exist when creating a file system.

For OLTP systems (see Figure 3-1 on page 86) we suggest having two LVs per
hdisk, one for data and one for backups. This allows parallelism when taking a
backup of the database by specifying all the backup file systems as targets.

For DW systems (see Figure 3-2 on page 87), we recommend having three LVs
per hdisk, one for data, one for backups and one for active logs/database
directory. This allows for data partition isolation.

We might need to create a striped LV if we want to increase capacity of the active

logging or archival logging file systems. We recommend you use the exact same
type of hdisk and use fine-grain stripping. We suggest a strip size of 32 K or 64 K.
Use the -S flag to indicate the stripSize when creating the LV.
mklv -t jfs2-y lvlogs -S 64K vg2 512GB hdisk7 hdisk8

3.6 File systems

A file system is a hierarchical structure (file tree) of files and directories. This type
of structure resembles an inverted tree with the roots at the top and branches at
the bottom. This file tree uses directories to organize data and programs into
groups, allowing the management of several directories and files at one time. A
file system resides on a single LV. Every file and directory belongs to a file
system within a LV.

Because of its structure, a few tasks are performed more efficiently on a file
system than on each directory within the file system. For example, you can back
up, move, or secure an entire file system. You can make a point-in-time image of
a JFS file system or a JFS2 file system, called a snapshot.

To be accessible, a file system must be mounted onto a directory mount point

using the mount command. When multiple file systems are mounted, a directory
structure is created that presents the image of a single file system. It is a
hierarchical structure with a single root. This structure includes the base file
systems and any file systems you create.

You can access both local and remote file systems using the mount command.
This makes the file system available for read and write access from your system.
Mounting or unmounting a file system usually requires system group
membership. File systems can be mounted automatically, if they are defined in
126 Best Practices for DB2 on AIX 6.1 for POWER Systems
the /etc/filesystemsfile. You can unmount a local or remote file system with the
unmount command, unless a user or process is currently accessing that file
system.

Both JFS and JFS2 file systems are built into the base operating system.
However it is not recommended to use JFS. You can also use other file systems
on AIX such as Veritas or GPFS, but for this Redbooks publication, we focus on
JFS2. This file system uses database journaling techniques to maintain its
structural consistency. It allows the file system log to be placed in the same
logical volume as the data, instead of allocating a separate logical volume for
logs for all file systems in the VG.

Because write operations are performed after logging of metadata has been
completed, write throughput is highly affected by where this logging is being
done. Therefore, we suggest using the INLINE logging capabilities. We place the
log in the LV with the JFS2 file system. We do not suggest setting any specific
size for the log. The INLINE log defaults to 0.4% of the LV size if logsize is not
specified. This is enough in most cases.

Another consideration for file systems are the mount options. They can be
specified in the crfs command with the -a option parameter or they can be
specified at mount time. We recommend specifying them at the crfs time to
ensure the proper options are used every time the file system is mounted. We
suggest the use of the release behind when read option (rbr option).

When sequential reading of a file in the file system is detected the real memory
pages used by the file are released after the pages are copied to internal buffers.
This solution addresses a scaling problem when performing sequential I/O on
large files whose pages are not reaccessed in the near future. When writing a
large file without using release-behind, writes go fast whenever there are
available pages on the free list.

When the number of pages drops to the value of the minfree parameter, the
virtual memory manager uses its Least Recently Used (LRU) algorithm to find
candidate pages for eviction. As part of this process, the virtual memory
manager needs to acquire a lock that is also being used for writing. This lock
contention might cause a sharp performance degradation.

A side effect of using the release-behind mechanism is an increase in CPU use

for the same read or write throughput rate compared to without using
release-behind. This is due to the work of freeing pages, which is normally
handled at a later time by the LRU daemon. Note that all file page accesses
result in disk I/O because file data is not cached by the virtual memory manager.
Chapter 3. Storage layout 127
 To create a file system, use the following commands:
mkdir -p <mountPoint>
crfs -v jfs2 -d <lvName> -m <mountPoint> -a logname=INLINE -a options=rbr -
A yes -p rw

3.7 Conclusion
This chapter described the storage principles guiding the storage layout, the
mapping of physical storage to logical storage, the layout of file systems, and
how DB2 uses those. We looked into OLTP and DW environments and what the
best practices are in each case. In general, we must design to spread our
hardware evenly, push as much functionality as possible down to the hardware,
have a simple LVM design and let DB2 handle the storage automatically. When
not sure how to configure certain parameters, the defaults are always a good
start. Then, proper monitoring helps to adjust. It is important to check the
performance as the system is being built as after the whole storage is laid out.
One change can mean a lot of re-work needed, in particular when it comes to the
lower layers of the design.
128 Best Practices for DB2 on AIX 6.1 for POWER Systems
 4

Chapter 4. Monitoring
Today’s environments range from stand-alone systems to complex combinations
of database servers and clients running on multiple platforms. In any type of
system, the common key for successful applications is performance. Although
performance might initially be good, as time goes on, the system might need to
serve more users, store more data, and process more complex queries.
Consequently, the increased load level on the system affects its performance.
This can be the time to upgrade to more powerful equipment. However, before
investing in equipment, you might be able to improve performance by monitoring
and tuning your environment.

This chapter provides an overview of the tasks involved in monitoring and tuning
DB2, AIX, and storage to obtain optimal performance and identify bottlenecks. It
describes how to monitor the system resource usage using AIX commands (such
as nmon, iostat, vmstat, and so forth). It also discusses the methods available to
monitor the database activity, using tools such as the Snapshot™ Monitor,
Activity Monitor and db2pd. Based on the information provided by these
commands and tools, you can make informed decisions about what actions need
to be taken to tune the database environment. Later in the chapter we include a
few scenarios in which we can use this monitoring information to determine
where the bottleneck can be.
© Copyright IBM Corp. 2010. All rights reserved. 129
This chapter has the following sections:
› “Understanding the system” on page 131
› “Benchmarking” on page 131
› “Determine the possible causes” on page 132
› “Planning monitoring and tuning” on page 133
› “Monitoring tools for DB2” on page 134
› “Monitoring enhancements in DB2 9.7” on page 149
› “Monitoring tools for AIX” on page 169
› “Monitoring scenarios” on page 185
130 Best Practices for DB2 on AIX 6.1 for POWER Systems
4.1 Understanding the system
It is important to have a good understanding of your system to figure out what
tools to use in the monitoring of bottlenecks and where they might lie. For
example, look at Figure 4-1, which details the several components that might
typically exist in today’s OLTP systems and data warehouses. After you know the
architecture, it becomes easier to perform calculated tests at the various levels to
determine where the problem can lie and what evidence to look for which
suggests a possible cause and consequently a solution.

Figure 4-1 Components involved

4.2 Benchmarking
After following the various best practices to configure the system, it is then
desirable to run a benchmark and establish the baseline. That is, for a given
workload X the response time for the various queries or jobs is Y. This applies to
both OLTP and data warehouses such that you can establish a baseline that
dictates how long a particular query or report takes. It is important to establish a
baseline and collect evidence of the system performing at this level, because if
the system performs at a substandard level in the future, you have a comparison
data that can aid in determining what can have lead to the performance
degradation.
Chapter 4. Monitoring 131
 The last point worth considering is of the analogy of comparing apples to apples,
For example, after migrating the various applications from the testing
environment to the production environment, the performance might be better or
worse. In such a situation a holistic picture must be taken to compare the
workload, configuration, hardware, and the like.

4.3 Determine the possible causes

There are many possible causes of performance problems. In the following
sections, we describe those most frequently encountered.

4.3.1 Poor application design

When you experience performance problems, in many cases these stem from
poor application design and inefficient programs. The underlying database,
operating system, or storage itself might not have any problems at all. For
example, SQL statements written inappropriately can degrade overall
performance, even though the system might be finely tuned and configured.
These type of problems are beyond the scope of this chapter.

4.3.2 Poor system and database design

Poor design of your system or databases can be also a reason for performance
problems. Inefficient disk storage layout, or failing to consider performance when
designing and implementing your database, degrades performance and can be
difficult to fix after your production system has been started. Chapter 2, “AIX
configuration” on page 35 and Chapter 3, “Storage layout” on page 85 describe
tips to take into consideration when designing disk storage layout and databases.

4.3.3 System resource shortages

System resource shortages can cause bottlenecks in your database system:
› CPU
Too many users, or running applications on a CPU, might cause system
degradation.
› Memory
Every process uses physical memory. If you have insufficient memory, you
might find that applications fails, or your system starts thrashing.
132 Best Practices for DB2 on AIX 6.1 for POWER Systems
 › Disk I/O
I/O performance can play an important part in a database system. Too much
activity on a single disk or I/O bus might cause performance problems.
› Network
Unless you are in a stand-alone environment, the network plays an important
role. If the network is too slow, this might appear to the client as a
performance problem at the database server.

It is these latter shortages and poor system and database design that are the
focus of this chapter.

4.4 Planning monitoring and tuning

When you carry out a performance monitoring and tuning project, the worst
possible approach is to change the value of many parameters without having any
idea of what is causing the performance problem. Barring miracles, performance
only gets worse, and you never know which parameter was the cause. So, even if
you are anxious for results, you benefit from following a more methodical
approach:
1. Find which applications or system resources are not performing well.
2. Measure the current performance and set the performance goal.
3. Monitor the system and identify where the bottleneck is.
4. Decide where you can afford to make trade-offs, and which resources can
bear an additional load.
5. Change only one performance parameter to relieve the bottleneck.
6. Execute the applications again to monitor your system, and check if the
performance meets your goal.
7. If the performance does not meet the goal, go back to step 3.

Note: Do not forget the Pareto's Principle, also known as the 80-20 rule. In our
context, 80% of the performance benefits can be the result of tuning 20% of
the various parameters.
Chapter 4. Monitoring 133
4.5 Monitoring tools for DB2
DB2 provides a suite of monitoring tools that can be effectively used to diagnose
performance problems. The collective power of AIX and DB2 diagnostics can be
used to resolve performance issues quickly. In this section, we discuss the most
useful DB2 performance diagnostics, ones that we can rely on to determine the
overall health of the database.

DB2 Monitoring Tools can be broadly classified into the following two categories:
› Point-in-time monitoring
› Traces

4.5.1 Point-in-time monitoring tools

The monitoring tools in this category are snapshot monitors, workload
management (WLM) aggregate functions starting DB2 Version 9.5, and
in-memory metrics in DB2 9.7. These tools provide us with a picture of the
database at a given point-in-time. They provide summarized data, where
counters, timers, and histograms maintain running totals of activity in the system.
It is strongly recommended to collect them on a regular basis for ongoing
monitoring. They are light in weight and can be useful in providing a picture of
database performance. If a performance problem arises, you can go back and
understand what has been happening and what has changed.

4.5.2 Traces
Tools in this category provide motion picture monitoring, which records the
execution of a series of individual activities. This is achieved with trace-like
mechanisms, such as event monitors (especially statement event monitors) and
WLM activity monitors. These tools provide a much more detailed picture of
system activity and therefore, produce huge volumes of data. This imposes a
much greater overhead on the system. They are more suitable for exception
monitoring when you need in-depth understanding of what is causing a
performance issue.

The recent trend in DB2 releases has been to move towards providing SQL
access to data generated by the monitoring tools. This makes monitoring more
manageable, as it allows you to redirect the output of snapshots and event
monitors back into tables. You can use the power of SQL to delve through the
historical monitoring data and get insights into your system performance.

Let us look at the most commonly used point-in-time diagnostics in DB2:

Snapshots and db2pd.
134 Best Practices for DB2 on AIX 6.1 for POWER Systems
4.5.3 Snapshots
Snapshots are the most commonly used performance diagnostics. DB2 provides
a set of administrative views that provides relational access to snapshot data.
Administrative views are available in SYSIBMADM schema. Examples of
administrative views are:
› SNAPDB,
› SNAPBP
› SNAPDBM
› SNAPAPPL
› BP_READIO

To get a complete list, use the following SQL:

select tabname from syscat.tables where tabschema='SYSIBMADM'

The instance-level monitor switches must be turned on if snapshot table

functions and administrative views are to access the data. For example,
DFT_MON_BUFPOOL enables collection of buffer pool monitoring data.

All snapshot elements provide good information but let us look at the most useful
ones: the Key Performance Indicators.

DB2 Key Performance Indicators (KPIs) and Rules of Thumb

(RoT)
Most of these KPIs are derived from database snapshot (GET SNAPSHOT FOR
DATABASE command). We use the columns of administrative view,
SYSIBMADM.SNAPDB for most of the metrics. If you are running in a
multi-partition environment, you must include DBPARTITIONNUM in your
monitoring SELECT statements, to distinguish the rows that you get back for
each partition.

Buffer Pool Hit Ratio

This metric measures percentage of reads that got satisfied in the buffer pool and
did not require physical reads. It is calculated as shown in Example 4-1.

Example 4-1 Buffer Pool Data/Index Hit Ratio

Data Hit Ratio

100 * (POOL_DATA_L_READS – POOL_DATA_P_READS)

/POOL_DATA_L_READS Index Hit Ratio

100 * (POOL_INDEX_L_READS – POOL_INDEX_P_READS) /POOL_INDEX_L_READS

Chapter 4. Monitoring 135

For OLTP, the RoT is to have Data Hit Ratio of at least 80% and Index Hit Ratio of
at least 95%.

Data warehouse workloads typically do large sequential table scans and

therefore do not have a high Buffer pool Hit Ratio. Buffer pool Temporary
Data/Index Hit Ratio is relevant for data warehouse workloads, as they are
characterized by complex queries that involve large sorts and hash joins. See
Example 4-2.

Example 4-2 Buffer pool Temporary Data/Index Hit Ratio

Temporary Data Hit Ratio

100 * (POOL_TEMP_DATA_L_READS - POOL_TEMP_DATA_P_READS) /

POOL_TEMP_DATA_L_READS

Temporary Index Hit Ratio

100 * (POOL_TEMP_INDEX_L_READS - POOL_TEMP_INDEX_P_READS) /

POOL_TEMP_INDEX_L_READS

Ideally we want this to be high to avoid temporary data physical I/O. However,
large scans and therefore, a low hit ratio is unavoidable.

Average Buffer pool I/O Response Time

This is a measure of the I/O response time and can be used in conjunction with
AIX monitoring tools to determine if there is a potential I/O Bottleneck. Keeping
modern storage subsystems in mind, average buffer pool read/write time is ~10
milliseconds.

Example 4-3 Overall Average Read Time

Overall Average Read Time(in ms)

POOL_READ_TIME /

(POOL_DATA_P_READS + POOL_INDEX_P_READS + POOL_TEMP_DATA_P_READS +

POOL_TEMP_INDEX_P_READS)

It can be useful to determine the average asynchronous and synchronous read

 time. Because OLTP is characterized by random reads, all the reads are
synchronous. Synchronous reads are done by the agents and significantly affect
the latency of transactions. See Example 4-4 on page 137.

136 Best Practices for DB2 on AIX 6.1 for POWER Systems
Example 4-4 Average Asynchronous/Synchronous Read Time

Average Asynchronous Read Time (in ms)

POOL_ASYNC_READ_TIME / (POOL_ASYNC_DATA_READS +
POOL_ASYNC_INDEX_READS)

Average Synchronous Read Time (in ms)

(POOL_READ_TIME -

POOL_ASYNC_READ_TIME)/
((POOL_DATA_P_READS + POOL_INDEX_P_READS +
POOL_TEMP_DATA_P_READS + POOL_TEMP_INDEX_P_READS) -

POOL_ASYNC_DATA_READS + POOL_ASYNC_INDEX_READS))

Similarly, you can get the average response time for buffer pool writes, as shown
in Example 4-5.

Example 4-5 Average Write Time

Average Write Time (in ms)
POOL_WRITE_TIME / (POOL_DATA_WRITES + POOL_INDEX_WRITES +

POOL_XDA_WRITES)

We can further breakup the Write Response time into Average Asynchronous
Write Time and Synchronous Write time, as shown in Example 4-6.

Example 4-6 Average Asynchronous/Synchronous Write Time

Average Asynchronous Write Time (in ms)

POOL_ASYNC_WRITE_TIME /
(POOL_ASYNC_DATA_WRITES +
POOL_ASYNC_INDEX_WRITES +
POOL_ASYNC_XDA_WRITES)

Average Synchronous Read Time(in ms)

(POOL_WRITE_TIME - POOL_ASYNC_WRITE_TIME) / ((POOL_DATA_WRITES +

POOL_INDEX_WRITES + POOL_XDA_WRITES) –
 (POOL_ASYNC_DATA_WRITES +
POOL_ASYNC_INDEX_WRITES +
POOL_ASYNC_XDA_WRITES)
)

Chapter 4. Monitoring 137

You can query the SYSIBMADM.BP_READ_IO and
SYSIBMADM.BP_WRITE_IO to obtain these I/O metrics for all buffer pools. See
Example 4-7.

Example 4-7 Query SYSIBMADM.BP_READ_IO & SYSIBMADM.BP_WRITE_IO

SELECT AVERAGE_READ_TIME_MS, AVERAGE_ASYNC_READ_TIME_MS,

AVERAGE_SYNC_READ_TIME_MS FROM SYSIBMADM.BP_READ_IO;

Transaction Log Response Time

Log response time significantly affects the latency of transactions in an OLTP
workload. It is therefore desirable to have low log I/O latency, not exceeding 10
milliseconds. Consider enabling storage write cache to reduce the log write
latency. See Example 4-8.

Example 4-8 Average Log Write Time

Average Log Write Time (in ms)

(LOG_WRITE_TIME_S*1000 + LOG_WRITE_TIME_NS/1000.0) / NUM_LOG_WRITE_IO

A high number of log buffer full, NUM_LOG_BUFFER_FULL, suggests that the

log buffer had to be flushed out to the disk as it became full. Increase the Log
Buffer Size to avoid this situation.

Page Cleaning
In OLTP, page cleaners play a crucial role. They asynchronously write dirty pages
to the disk. This ensures that there are clean slots available for use when a new
page needs to be read into the buffer pool. We want to make sure that most of
the writes are being done asynchronously by the page cleaners. If not, it is the
agent that is doing the writes which is expensive. See Example 4-9.

Example 4-9 Page Cleaning Ratio

Page Cleaning Ratio

100 * (POOL_ASYNC_DATA_WRITES + POOL_ASYNC_INDEX_WRITES +

POOL_ASYNC_XDA_WRITES) / (POOL_DATA_WRITES + POOL_INDEX_WRITES +

POOL_XDA_WRITES

A 95% page cleaning ratio is considered good.

138 Best Practices for DB2 on AIX 6.1 for POWER Systems
No victim buffers available
POOL_NO_VICTIM_BUFFER counts the number of times the free page list was
empty when a clean page was needed. Consider this as an early warning of
inefficient cleaning. If this number gets large, try increasing the number of
cleaners, or decreasing CHNGPGS_THRES or SOFTMAX to trigger cleaning at
a higher frequency

Dirty page steals

Having no victim buffers available and dirty page steals go hand-in-hand, as dirty
steals are indicative of no suitable victim pages being found. No victim buffers
available is more sensitive than Dirty Page Steals.

POOL_DRTY_PG_STEAL_CLNS counts the number of times a dirty page has to

be cleaned before a read. This results in synchronous writes that can be
detrimental for throughput. A dirty page steal rate greater than 1 per 100
transactions can be considered non-trivial. This counter is low, ideally zero.

Dirty page threshold / LSN Gap Triggers

There are three ways to trigger cleaners. We need cleaners to run to have free
pages, and we want to avoid dirty page steals. Dirty page threshold triggers,
POOL_DRTY_PG_THRSH_CLNS and LSN gap triggers,
POOL_LSN_GAP_CLNS are considered good cleaner triggers. If their number
gets too large, we might be over-cleaning. Increasing CHNGPGS_THRESH or
SOFTMAX decreases the cleaner frequency.

Consider enabling DB2 Alternate Page Cleaning (db2set

DB2_USE_ALTERNATE_PAGE_CLEANING=ON). This proactively cleans the
dirty pages from the buffer pool such that there is a constant writing of dirty
pages that can alleviate the I/O spikes associated with the triggers.

Prefetch Ratio
This represents the percentage of asynchronous reads done by the prefetcher
and is particularly relevant for sequential scans. RoT is to keep this close to
100%. Prefetch Ratio is high for data warehouse workloads
(POOL_ASYNC_DATA_READS + POOL_ASYNC_INDEX_READS) /
(POOL_DATA_P_READS + POOL_INDEX_P_READS +
POOL_TEMP_DATA_P_READS + POOL_TEMP_INDEX_P_READS)
Chapter 4. Monitoring 139
Rows Read/Row Selected Ratio
This is a useful metric. It determines how many rows had to be read to return the
selected rows. A high ratio is an indication of large table scans and indicates
possibility of creating indexes. This is particularly relevant for OLTP, where we do
not want this ratio to be greater than 10–20. You can drill down to Statement
Snapshots to find the culprit SQLs. Look for SQLs that have Rows Read much
higher than the Number of Executions.

ROWS_READ / ROWS_SELECTED

Note that repeated in-memory table scans can also consume significant CPU.

Package Cache Hit Ratio

This determines the fraction of SQL prepares that cannot be satisfied by dynamic
SQL cache. It is calculated as:

(1-(pkg_cache_inserts / pkg_cache_lookups))*100%

A low package cache ratio is either because we are re-compiling the same SQL
statement with other literal values or because of low package cache size. RoT is
to have it close to 100% at steady state.

Sort Metrics
When using private sort memory, SHEAPTHRES is a soft limit. If the
SHEAPTHRES limit is exceeded, new sorts get significantly less than their
optimal amount of memory, degrading query performance. Database Manager
Snapshot can be used to determine if we exceeded the SHEAPTHRES. If the
Private Sort heap high water mark is greater than SHEAPTHRES, consider
increasing SHEAPTHRES.

Another important sort metrics is the number of Sort overflows,

SORT_OVERFLOWS. It represents how many sorts spilled to the disk. It is good
to have near to 0 sort overflows for OLTP workloads.

Lock Metrics
Contention issues, deadlocks, and lock timeouts can be attributed to poor
application design. They can drastically affect the response time and throughput.
› LOCK_ESCALS
› LOCK_TIMEOUTS
› DEADLOCKS

Lock escalations (measured by LOCK_ESCALS) happen when we run out of

space on the locklist. Increase LOCKLIST to avoid escalations. Lock timeouts
are not necessarily fatal but they do decrease the concurrency. LOCK_TIMEOUT
is set to catch problem situations. You can further delve into an application
140 Best Practices for DB2 on AIX 6.1 for POWER Systems
snapshot to determine the percentage of applications in Lock-wait state to
determine if there are any locking issues.

A deadlock is created when two applications are each locking data needed by
the other, resulting in a situation when neither application can continue to
process.

The best tool for collecting information about a deadlock is a detailed deadlock
event monitor, which must be defined and enabled before the problem occurs. If
you do not have the default deadlock event monitor running, you must create and
enable a detailed deadlock event monitor.

For information about the default deadlock monitor, see the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9/index.jsp?topic=/com
.ibm.db2.udb.admin.doc/doc/c0005419.htm
Dynamic SQL Metrics
Dynamic SQL Snapshot gives us good insights into all dynamic SQLs processed.
This helps us determine the hottest statements in the database and see if there
is an opportunity to improve their performance. You can query the
SYSIBMADM.SNAPDYN_SQL to access dynamic SQL snapshot data.

NUM_EXECUTIONS, NUM_COMPILATIONS, ROWS_READ,

TOTAL_SORT_TIME, STMT_SORTS and hit ratios are the key metrics we look
at while determining the performance characteristics of dynamic SQLs.

There are instances when you notice a big difference between the execution time
and User + System CPU Time.
TOTAL_EXEC_TIME - (TOTAL_USR_CPU_TIME + TOTAL_SYS_CPU_TIME)

This is because Total Execution time includes all the white space between open
and close (fetches, application code, I/O wait and network wait). It is not a good
indicator of real work being done.

You can query the SYSIBMADM. TOP_DYNAMIC_SQL view to find the hottest
statements in the database and sort the result set by other metrics (for example,
number of executions, average execution time and number of sorts per
execution).

Important: An important thing to keep in mind when you compare two

snapshots is to normalize the metrics based on the amount of activity
recorded during the monitoring period. The number of commit statements,
COMMIT_SQL_STMTS, is the metric that reflects the number of transactions
and is a good indicator of the amount of activity in the database. This gives
what we call in the performance world an apples-to-apples comparison.
Chapter 4. Monitoring 141
4.5.4 db2top
Most entries in snapshots are cumulative values and show the condition of the
system at a point in time. Manual work is needed to get a delta value for each
entry from one snapshot to the next.

The db2top is a tool that uses DB2 snapshot monitoring APIs to retrieve
information about the database system. It is used to calculate the delta values for
those snapshot entries in real time. This tool provides a GUI under a command
line mode, so that users can get a better understanding while reading each entry.
This tool also integrates multiple types of DB2 snapshots, categorizes them, and
presents them in other windows for the GUI environment.

db2top has been included in the following DB2 versions and later:
› DB2 UDB v8.2 Fixpak 17 and later
› DB2 UDB v9.1 Fixpak 6 and later
› DB2 UDB v9.5 Fixpak 2 and later
› DB2 UDB v9.7 GA and later

The following Web page provides an in-depth discussion on the db2top tool:
http://www.ibm.com/developerworks/data/library/techarticle/dm-0812wang/
index.html

4.5.5 db2pd
db2pd is a standalone utility shipped with DB2 starting with V8.2. It provides a
non-intrusive method to view database progress and potential problems. You can
use the power of db2pd to get insights into the DB2 engine. Some examples of
information provided by db2pd are:
› Memory usage
› Agents
› Applications
› Locks
› Buffer pools
› Dynamic package cache
› Static package cache
› Catalog cache
› Logs
› Table and index statistics

In this section, we discuss db2pd options that you can use to diagnose
performance problems.
142 Best Practices for DB2 on AIX 6.1 for POWER Systems
db2pd –edus
Multi-threaded architecture has been introduced on UNIX starting with DB2 9.5.
Prior to DB2 9.5, you can see all active DB2 processes using the ps command on
UNIX. Starting with DB2 9.5, the ps command only lists the parent process,
db2sysc. You can use db2pd –edus to list all DB2 threads, along with their CPU
use. It can be useful to determine which DB2 thread is behind a CPU bottleneck.
See Example 4-10.

Example 4-10 db2pd –edus

Database Partition 0 -- Active -- Up 0 days 00:47:51 List of
all EDUs for database partition 0
db2sysc PID: 13360
db2wdog PID: 13358
db2acd PID: 13374

EDU ID TID Kernel TID EDU Name USR (s) SYS (s)
219 4791480614329 13976 db2pclnr (DTW) 0.000000 0.000000
6
218 47914814531904 13975 db2pclnr (DTW) 0.000000 0.000000
217 47914822920512 13944 db2dlock (DTW) 0.000000 0.000000
216 47914831309120 13943 db2lfr (DTW) 0.000000 0.000000
215 4791494874963 13942 db2loggw (DTW) 0.760000 2.450000
2
214 47914827114816 13925 db2loggr (DTW) 0.090000 0.120000
213 4792639919955 13825 db2stmm (DTW) 0.040000 0.010000
2
191 4791483550342 13802 db2agent (DTW) 45.280000 7.810000
4
190 4791483969772 13801 db2agent (DTW) 46.310000 7.610000
8
189 4791484389203 13800 db2agent (DTW) 44.950000 7.600000
2
188 4791484808633 13799 db2agent (DTW) 46.000000 7.700000
6

You can use the AIX ps command, ps -mo THREAD -p <db2sysc pid>, as well to
get details about the EDU threads.

db2pd –memsets and –mempools

These can give an insight into memory usage of DB2 and can be useful to look at
when you run into memory bottleneck. They can also be used to detect memory
leaks in DB2, which are rare.

-memsets is used to gain a quick, detailed view of how much memory each DB2
memory set is using. See Example 4-11 on page 144.

–mempools is used to drill down further into the memory pools that constitute a
memory set.
Chapter 4. Monitoring 143
Example 4-11 Memory set
db2pd -memset
Database Partition 0 -- Active -- Up 0 days 10:07:04
Memory Sets:
Name Address Id Size(Kb) Key
DBP Type Unrsv(Kb) Used(Kb) HWM(Kb) Cmt(Kb)
Uncmt(Kb)
DBMS 0x0780000000000000 28311566 36288 0x931FF261 0
0 0 14336 17024 17024 19264
FMP 0x0780000010000000 223346762 22592 0x0 0
0 2 0 576 22592 0
Trace 0x0770000000000000 362807300 137550 0x931FF274 0
-1 0 137550 0 137550 0

In Example 4-11:
› Name is the name of memory set.

› Addressis the address of the memory set.

› Id is the memory set identifier.

› Size(Kb)is the size of memory set in kilobytes.

› Keyis the memory set key (for Unix-based systems only).

› DBPis the database partition server that owns the memory set.

› Type is the type of memory set.

› Unrsv(Kb)is the memory not reserved for any particular pool. Any pool in the
set can use this memory if needed.
› Used(Kb)shows memory currently allocated to memory pools.
› Cmt(Kb)shows all memory that has been committed by the DB2 database
and occupies physical RAM, paging space or both.
› Uncmt(Kb)shows memory not currently being used and marked by the DB2
database to be uncommitted.
144 Best Practices for DB2 on AIX 6.1 for POWER Systems
The DBMS Memory set corresponds to INSTANCE_MEMORY dbm cfg. Let us
delve further into the memory pools owned by the DBMS memory set. See
Example 4-12.

Example 4-12 db2pd -mempools

db2pd -mempools
Database Partition 0 -- Active -- Up 0 days 10:17:47
Memory Pools:
Address MemSet PoolName Id Overhead LogSz
LogUpBnd LogHWM PhySz PhyUpBnd PhyHWM Bnd BlkCnt
CfgParm
0x07800000000012A DB fcm 74 0 0
8 MS
1931054 0 0 1966080 0 Ovf 0
n/a
0x0780000000001160 DB fcmsess 77 65376 1401568
MS
1687552 1401568 1572864 1703936 1572864 Ovf 3
n/a
0x0780000000001018 DB fcmchan 79 65376 259584
MS
507904 259584 393216 524288 393216 Ovf 3
n/a
0x0780000000000ED DB fcmbp 13 65376 656896
0 MS
925696 656896 851968 983040 851968 Ovf 3
n/a
0x0780000000000D8 DB fcmctl 73 111872 1594237
8 MS
8675472 1769472 8716288 1769472 Ovf 11
159423
7
n/a
0x0780000000000C4 DB monh 11 0 3664
0 MS
368640 144759 196608 393216 327680 Ovf 4
MON_HEAP_SZ
0x0780000000000AF DB resynch 62 41216 155320
8 MS
2752512 155320 262144 2752512 262144 Ovf 2
n/a
0x07800000000009B DB apmh 70 4512 1757972
0 MS
5898240 1915716 1900544 5898240 1966080 Ovf 86
n/a
0x078000000000086 DB kerh 52 0 1564968
8 MS
4128768 1951064 1900544 4128768 2228224 Ovf 167
n/a
0x078000000000072 DB bsuh 71 0 61369
0 MS
13828096 262144 13828096 2490368 Ovf 19
227636
2
n/a
0x07800000000005D8 DBMS sqlch 50 0 2560840
2621440 2560840 2621440 2621440 2621440 Ovf 203
n/a

Chapter 4. Monitoring 145

 0x0780000000000490 krcbh 69 0 145272
DBMS
131072 145624 196608 131072 196608 Ovf 11
n/a
0x0780000000000348 eduah 72 65440 2621384
DBMS
2621408 2621384 2686976 2621440 2686976 Ovf 1
n/a
0x0780000010000348 FMP undefh 59 32000 491600
22971520 491600 524288 23003136 524288 Phy 4
n/a

› The (Size-Unrsv) of a memory set is equal to the sum of PhyUpBnd of all

memory pools owned by the memory set.
› The Used size of a memory set is almost equal to sum of PhySz of memory
pools in the memory set.

You can further delve into the logical memory of a memory pool using the db2pd
–memblocks option. You can view this information at the database level as well to
get overview of the database memory set and memory pools

If the Used size of memory set is much higher than the sum of the physical size
of memory pool, it might indicate a memory leak.

db2pd -tcbstats
This option displays information about the size and activity of tables. The Scan
column can be particularly useful in determining table scans. A steadily
increasing Scan value on a table can indicate a poor plan and a potential
requirement to create an index. In this example, there have been 30419 table
scans on the LAST_TRADE table. Note that although LAST_TRADE is a small
table containing only 643 rows, repeated in-memory table scans can lead to a
potential CPU bottleneck. See Example 4-13.

Example 4-13 db2pd -tcbstat

TableName SchemaNm Scans RTSUDI PgReorgs

NoChgUpdts UD Updates Deletes OvFlReads OvFlCrtes
Reads I FscrUpdates Inserts
RowsComp RowsUncomp CCLogReads BytesSaved
StoreBytes 4769 0 0
LAST_TRADE SW 30419 4769
20789015 0 0 4769 0 0 0 0 0
256621 - -
146 Best Practices for DB2 on AIX 6.1 for POWER Systems
 db2pd –activestatements
This option indicates the dynamic statements agents are currently executing. You
can use this to determine the most active statements in your database. In
Example 4-14, statement with Anchor ID = 600 seems to be fairly active. db2pd
–dynamic can help get the statement text corresponding to Anchor ID=600. See
Example 4-14.

Example 4-14 db2pd – activestatements

Address AppHandl [nod-index] UOW- StmtID StmtUID

EffISO ID EffLockTOut EffDegree EntryTime AnchI L
D StartTime
0x00002B9674BF4E2 138 [000-00138] 1907 15 928 1 0
0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674BFABE 118 [000-00118] 1956 19 600 1 0
0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674C37D 131 [000-00131] 1946 38 701 1 0
80
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674CB008 144 [000-00144] 1920 23 600 1 0
0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674CD9C 111 [000-00111] 1886 31 600 1 0
C0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674C55E4 124 [000-00124] 1901 19 600 1 0
0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674C61FC 130 [000-00130] 1946 13 1011 1 0
0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674CFA3 143 [000-00143] 1885 11 600 1 0
A0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674C31FC 110 [000-00110] 1924 16 760 1 0
0
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674D0BC 123 [000-00123] 1898 15 600 1 0
00
-1 0 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
d
0x00002B9674C6DB 136 [000-00136] 1855 15 600 1 0
40 Wed Dec 2 06:33:20 We Dec 2 06:33:20 W
-1 0
d

Chapter 4. Monitoring 147

 db2pd –dynamic
db2pd –dynamic helps identify the statement text with anchor ID = 600. NumRef
indicates how many times a statement has been referenced and hence, indicates
how active the dynamic statement. See Example 4-15.

Example 4-15 db2pd –dynamic

Database Partition 0 -- Database DTW -- Active -- Up 0 days 00:04:12

Dynamic Cache:
Current Memory Used 406678
4
Total Heap Size 406847
4
Cache Overflow Flag 0
Number of References 954293
Number of Statement Inserts 12343
Number of Statement Deletes 12072
Number of Variation Inserts 12399
Number of Statements 271

Dynamic SQL Statements:

Address AnchID StmtUID NumEnv NumVar NumR NumExe Text

ef
0x00002B9674E4F3A0 14 1 1 1 24009 24007
Insert into HISTORY (H_W_ID, H_D_ID, H_C_W_ID, H_C_D_ID, H_C_ID, H_DATE, H_AMOUNT,
H_DATA) values (?, ?, ?, ?, ?, ?, ?, ?)

0x00002B9674E4FEC0 600 1 1 1 250037 250037

Select S_QUANTITY, S_DIST_01,S_DIST_02, S_DIST_03, S_DIST_04, S_DIST_05, S_DIST_06,
S_DIST_07, S_DIST_08, S_DIST_09, S_DIST_10, S_YTD, S_ORDER_CNT, S_REMOTE_CNT, S_DATA
from STOCK where S_W_ID = ?and S_I_ID = ?
148 Best Practices for DB2 on AIX 6.1 for POWER Systems
 db2pd –static
Snapshots only monitor the dynamic SQL statements. Although we can use the
statement event monitor to track Static SQLs, they generate huge volumes of
data, making them impractical for on-going monitoring. db2pd –static is a handy
option to determine the active Static SQL statements.

See Example 4-16. Section 9 of package PAYS is active as it is being referenced

by 15 concurrent users. You can query SYSCAT.STATEMENTS to determine the
query that correspond to Section 9 of PAYS package.

Example 4-16 db2pd – static

Database Partition 0 -- Database DTW -- Active -- Up 0 days 00:06:32 Static

Cache:
Current Memory Used 232831
Total Heap Size 4068474
Cache Overflow Flag 0
Number of References 716105
:Sections:
Address Schema UniqueID SecNoNumRef UseCount StmtType Cursor
PkgName
W-Hld
0x3FB55138 SW PAYS JAxPMQCV 7 1604628 0 6
NO
0x3FB5527C SW PAYS JAxPMQCV 8 1604626 0 4
NO
0x3FB553C0 SW PAYS JAxPMQCV 9 1604625 15 4
NO

This shows how db2pd can be effective at times in diagnosing performance

issues.

4.6 Monitoring enhancements in DB2 9.7

DB2 9.7 introduces several enhancements in the area of database monitoring.
The primary goal of these enhancements is to reduce the time and effort
required to detect and resolve common database performance problems. Major
enhancements have been made to improve the diagnostics as well as to reduce
the performance impact of monitoring, both in the area of point-in-time
monitoring and event monitors. We discuss the changes in point-in-time
monitoring in the following sections.
Chapter 4. Monitoring 149
4.6.1 In-memory metrics: New monitoring infrastructure
In DB2 9.7, we begin to move away from traditional snapshot towards a new
light-weight monitoring infrastructure, In-memory metrics. Key highlights of the
new infrastructure are:
› A comprehensive set of Time Spent in DB2 metrics that tells you how and
where DB2 is spending time.
› Reduced performance impact of collecting and querying monitoring data.

Note: Tests done in the lab on an OLTP workload suggest that the
performance overhead of collecting Snapshots (all switches ON) is 6%,
while it is only 3% with In-memory metrics.

› Better control over the granularity of monitoring. You can enable monitoring
only for specific service classes.
› A common interface through Package Cache for monitoring both Static and
Dynamic SQL

How to use the new monitoring infrastructure

Data for monitoring elements is accumulated and available for querying along
multiple perspectives:
› System level
› Activity level
› Database object level

System level monitoring

System level monitoring reflects the effort DB2 spent in processing application
requests. It allows you to determine what the system is doing as a whole, as well
as for particular categories and subsets of workload.

Request metrics are accumulated in-memory and reported at the following

points:
› Service Subclass
› Connection
› Workload Definition
› Unit of Work
150 Best Practices for DB2 on AIX 6.1 for POWER Systems
As shown in Figure 4-2, a database request is processed by an agent. The agent
gathers the in-memory metrics and rolls up the metrics to other aggregation
points at regular intervals. (Transaction boundary or every 10 seconds, whichever
comes first).

Figure 4-2 How metrics are gathered

Monitoring can be enabled at two levels:

› Database level
UPDATE DB CFG FOR <DBNAME> USING MON_REQ_METRICS BASE
› Service Super Class level
CREATE SERVICE CLASS <SERVICE-CLASS-NAME> COLLECT REQUEST DATA BASE

You can also monitor only a set of workloads by enabling monitoring only for the
service class to which the workload maps. The database level decides the
minimum level of monitoring. For example, if MON_REQ_METRICS=BASE,
then request metrics are collected regardless of service class setting.
MON_REQ_METRICS is set to BASE for new databases and to NONE for
migrated database.

System in-memory metrics are exposed through the following table functions:
› MON_GET_UNIT_OF_WORK, MON_GET_UNIT_OF_WORK_DETAILS
› MON_GET_WORKLOAD, MON_GET_WORKLOAD_DETAILS
› MON_GET_CONNECTION, MON_GET_CONNECTION_DETAILS
› MON_GET_SERVICE_SUBCLASS,
MON_GET_SERVICE_SUBCLASS_DETAILS
Chapter 4. Monitoring 151
Let us see a few examples.

List the active connections where we are spending most of our time

We query the MON_GET_CONNECTION table function to get a high level view

of where we are spending time. It accepts two arguments:
› Application_handle
› Member

If the application handle is NULL, details are returned for all connections.
Member is used to specify the node in DPF environment. Specify -1 for the
current node and -2 for all nodes.

TOTAL_RQST_TIME is the total time DB2 spent for all requests for an
application. ACT_COMPLETED_TOTAL represents the number of activities
(SQL statements) executed by the application. See Example 4-17.

Example 4-17 List the active connection

SELECT APPLICATION_HANDLE, ACT_COMPLETED_TOTAL,

TOTAL_RQST_TIME FROM TABLE(MON_GET_CONNECTION(CAST(NULL AS
BIGINT), -2))
ORDER BY TOTAL_RQST_TIME DESC
APPLICATION_HANDLE ACT_COMPLETED_TOTAL TOTAL_RQST_TIME
-------------------- -------------------- --------------------
11 128279 198963
14 128377 198813
9 128580 198615
13 128491 198486
12 126753 198403
8 125708 198371
7 126950 197918
10 127289 197889
37 1 58127
15 10 33079

In the output, application handle = 37 has executed only one activity, while other
applications have completed relatively high number of activities.
152 Best Practices for DB2 on AIX 6.1 for POWER Systems
What are the transactions (UOW) the connections are currently executing?

We can drill down to the UOW level through MON_GET_UNIT_OF_WORK to

determine this. MON_GET_UNIT_OF_WORK accepts application_id and
member as arguments. See Example 4-18.

Example 4-18 Drill down to the UOW level

SELECT APPLICATION_HANDLE, UOW_ID,

TOTAL_RQST_TIME, WORKLOAD_OCCURRENCE_STATE,
UOW_START_TIME FROM
TABLE(MON_GET_UNIT_OF_WORK(NULL,-2))
WHERE
APPLICATION_HANDLE IN (
SELECT
APPLICATION_HANDLE
FROM TABLE(MON_GET_CONNECTION(CAST(NULL AS BIGINT), -2) )
AS T ORDER BY TOTAL_RQST_TIME DESC FETCH FIRST 10 ROWS
ONLY)
ORDER BY TOTAL_RQST_TIME DESC

APPLICATION_HANDLE UOW_ID
TOTAL_RQST_TIME
WORKLOAD_OCCURRENCE_STATE
UOW_START_TIME
-------------------- ----------- --------------------
-------------------------------- --------------------------
37 1 65444
UOWEXEC 2009-12-07-08.41.42.321700
13 5455 0
UOWEXEC 2009-12-07-08.43.27.979914
12 5500 0
UOWEXEC 2009-12-07-08.43.28.009260
11 5457 0
UOWEXEC 2009-12-07-08.43.28.030864
10 5377 0
UOWEXEC 2009-12-07-08.43.28.008118
9 5445 0
UOWEXEC 2009-12-07-08.43.28.020324
 15 12 0
UOWEXEC 2009-12-07-08.43.27.943657
8 5509 0
UOWWAIT 2009-12-07-08.43.27.959482
14 5417 0
UOWEXEC 2009-12-07-08.43.28.025917
7 5480 0
UOWEXEC 2009-12-07-08.43.28.023564
10 record(s) selected.

Chapter 4. Monitoring 153

 Application 37 look suspicious. It seems to be executing a long running
transaction. Monitoring the UOWs at regular intervals consistently shows
UOW_ID=1 from APPLICATION HANDLE=37 at the top of the list. For other
applications, the UOW IDs in a state of execution are constantly changing. We
can further drill down to the activity level to determine what the UOW is
executing. We discuss this in “Activity level monitoring” on page 155.

If you have configured WLM, you can obtain metrics at the service class level and
workload level. Let us see an example:
› Determine the number of completed activities and CPU Time consumed by
each service class
We can query MON_GET_SERVICE_SUBCLASS. See Example 4-19.

Example 4-19 Query the MON_GET_SERVICE_SUBCLASS

SELECT varchar(service_superclass_name,30) as service_superclass, varchar(service_subclass_name,30) as

service_subclass,
total_cpu_time as total_cpu,
act_completed_total as SQL_EXECUTED
FROM
TABLE(MON_GET_SERVICE_SUBCLASS('','',-2)) AS t
ORDER BY total_cpu desc

SERVICE_SUPERCLASS SERVICE_SUBCLASS
TOTAL_CPU SQL_EXECUTED
------------------------------
--------------------------------------------------------------- FINANCE
TRANSACTCLASS
5673098558
43994853
FINANCE REPORTCLASS 1341307246
66
SYSDEFAULTUSERCLASS SYSDEFAULTSUBCLASS 628763
13
SYSDEFAULTMAINTENANCECLASS SYSDEFAULTSUBCLASS 113371
0
SRVCLASS SYSDEFAULTSUBCLASS 0
0
SYSDEFAULTSYSTEMCLASS SYSDEFAULTSUBCLASS 0
0

6 record(s) selected.
 If you want to view the metrics of a particular class, specify the service super
class name and service subclass name as first two arguments of
MON_GET_SERVICE_SUBCLASS.

154 Best Practices for DB2 on AIX 6.1 for POWER Systems
It is good practice to configure WLM such that there is one-to-one mapping
between an application and a workload/service sub class. This enables you to
determine the performance characteristics of an application by looking at service
class/workload monitoring metrics.

Activity level monitoring

This provides reporting from the perspective of the individual activities processed
by DB2, such as the execution of individual SQL statements. For SQL
statements, the term activity refers only to the execution of the section for that
statement. Metrics for other requests or processing that is performed before or
after section execution are not included in the activity metrics. See Figure 4-3.

Figure 4-3 Activity level monitoring

A request comes into an agent and is processed. If the request is related to an

activity, the agent gathers the metrics from the start of activity execution and, at
regular intervals, aggregates them in the activity control block. When the activity
completes, those activity execution metrics are propagated to the package cache
and aggregated under the specific cached section that was executed (static and
dynamic SQL).

Monitoring can be enabled at two levels:

› Database level
UPDATE DB CFG FOR <DBNAME> USING MON_ACT_METRICS
BASE
› Service Super Class level
Chapter 4. Monitoring 155
 CREATE SERVICE CLASS <SERVICE-CLASS-NAME> COLLECT ACTIVITY
METRICS BASE

The database level decides the minimum level of monitoring. For example, if
MON_ ACT_METRICS=BASE, then in-memory metrics are collected regardless
of service class setting. MON_ ACT_METRCIS is set to BASE for new databases
and to NONE for migrated database. Activity metrics are exposed through:
› MON_GET_PKG_CACHE_STMT (Both static and dynamic SQL)
› MON_GET_ACTIVITY_DETAILS (XML)

MON_GET_ACTIVITY_DETAILS returns data about the individual activities in

progress when the table function is called. Data is returned in XML format.

MON_GET_PKG_CACHE_STMT returns data for an individual SQL statement

section aggregated over all executions of the section. Data is returned in a
relational form.

Continuing our previous example, we want to find details about the SQL
statements that the problematic UOW is currently executing.

MON_GET_ACTIVITY_DETAILS returns metrics about a specific activity

identified by its application handle, unit of work ID, and activity ID

We can determine the activity_id using the WLM Table function,

WLM_GET_WORKLOAD_OCCURRENCE_ACTIVITIES_V97, as shown in
Figure 4-20 on page 182.

Example 4-20 Determine the activity_id

SELECT application_handle, activity_id, uow_id, local_start_time
FROM TABLE (WLM_GET_WORKLOAD_OCCURRENCE_ACTIVITIES_V97(37, -1) ) AS T

APPLICATION_HANDLE ACTIVITY_ID UOW_ID LOCAL_START_TIME

-------------------- ----------- ----------- -------------------------- 37 2

1 2009-11-23-12.26.10.824045
156 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Next, we use the activity_id to determine which SQL statement corresponds to
this activity and to determine where it is spending time. See Example 4-21.

Example 4-21 Determine which SQL statement corresponds to this activity

SELECT actmetrics.application_handle,
actmetrics.activity_id,
actmetrics.uow_id,
varchar(actmetrics.stmt_text, 50) as stmt_text,
actmetrics.total_act_time,
actmetrics.total_act_wait_time,
actmetrics.total_act_time,
actmetrics.total_act_wait_time,
CASE WHEN actmetrics.total_act_time > 0
THEN DEC((
FLOAT(actmetrics.total_act_wait_time) /
FLOAT(actmetrics.total_act_time)) * 100, 5, 2)
ELSE NULL
END AS PERCENTAGE_WAIT_TIME
FROM TABLE(MON_GET_ACTIVITY_DETAILS(37, 1 , 2, -2)) AS ACTDETAILS,
XMLTABLE (XMLNAMESPACES( DEFAULT 'http://www.ibm.com/xmlns/prod/db2/mon'),
'$actmetrics/db2_activity_details'
PASSING XMLPARSE(DOCUMENT ACTDETAILS.DETAILS) as "actmetrics"
COLUMNS "APPLICATION_HANDLE" INTEGER PATH
'application_handle', "ACTIVITY_ID" INTEGER PATH 'activity_id',
"UOW_ID" INTEGER PATH 'uow_id',
"STMT_TEXT" VARCHAR(1024) PATH 'stmt_text',
"TOTAL_ACT_TIME" INTEGER PATH 'activity_metrics/total_act_time',
"TOTAL_ACT_WAIT_TIME" INTEGER PATH 'activity_metrics/total_act_wait_time'
) AS ACTMETRICS;

APPLICATION_HANDLE ACTIVITY_ID UOW_ID STMT_TEXT

TOTAL_ACT_TIME TOTAL_ACT_WAIT_TIME
PERCENTAGE_WAIT_TIME
------------------ ----------- -----------
-------------------------------------------------- -------------- -------------------
--------------------
37 2 1 select * from sw.order_line
142045 121702 85.67
1 record(s) selected.
Chapter 4. Monitoring 157
 This shows that the query corresponding to the long-running UOW is spending
85% of its time in wait. MON_GET_ACTIVITY_DETAILS only gives information
about activities that are currently in progress. To determine your hottest SQL
statements over a period of time, you can query
MON_GET_PKG_CACHE_STMT. It is similar to Dynamic SQL Snapshot but
with an added advantage: you can monitor both Static and Dynamic SQLs!

MON_GET_ PKG_CACHE_STMT accepts four arguments:

› Section: Type ‘S’ for Static SQLs and ‘D’ for Dynamic SQLs.
› Executable ID: Specifies a unique identifier for section in package cache.
› Search_args: An optional input parameter of type CLOB(1K), that allows you
to specify one or more optional search argument strings.
› Member: Node number.

What are my top 5 hottest SQL statements, sorted by Rows Read?:

If NULL is specified as Executable_ID, we get all rows. To filter rows, we select

the Executable IDs of the top 10 SQLs (highlighted in blue) and use them as an
argument to the MON_GET_PKG_CACHE_STMT

Unlike Dynamic SQL Snapshot, we get the Rows Returned that can quickly help
us determine the SQL statements that have a high Rows Read/Rows Returned
Ratio. See Example 4-22.

Example 4-22 What are my top 5 hottest SQL statements, sorted by Rows Read?
SELECT
NUM_EXECUTIONS, STMT_EXEC_TIME,
ROWS_READ, ROWS_RETURNED,
SUBSTR(STMT_TEXT,1, 200) AS STMT_TEXT
FROM
(SELECT EXECUTABLE_ID
FROM TABLE ( MON_GET_PKG_CACHE_STMT (NULL, NULL, NULL, -
2) ) ORDER BY ROWS_READ DESC FETCH FIRST 5 ROWS ONLY ) T1,
TABLE (MON_GET_PKG_CACHE_STMT(NULL,T1.EXECUTABLE_ID ,NULL,-2));

ROWS_READ ROWS_RETURNED STMT_TEXT

-----------------------------------------------------------------------
70794652 350 select nation, o_year, sum(amount) as s
57705617 100 select s_name, count(*) as numwait from
56182450 15 select n_name, sum(l_extendedprice *
49149656 200 select c_name, c_custkey, o_orderkey,
48449096 4 select l_shipmode, sum(case when i
5 record(s) selected.
 All the queries are aggregate queries, doing huge table scans.

158 Best Practices for DB2 on AIX 6.1 for POWER Systems
Database object monitoring
This provides reporting from the perspective of operations performed on
particular database objects (table, index, buffer pool, table space and container).
It provides a complementary perspective of database operation to the workload,
allowing the user to pinpoint issues from a data object centric perspective rather
than a workload centric perspective. See Figure 4-4.

Figure 4-4 Database object monitoring

When agents perform work on the system and gather low-level monitoring
metrics, relevant metrics are propagated to accumulation points in in-memory
data objects as well as the aforementioned accumulation points in the workload
framework.

Database object monitoring can be enabled only at the database level.

DB2 UPDATE DB CFG FOR <DBNAME> USING MON_OBJ_METRICS BASE

Database objects metrics are exposed through:

› MON_GET_TABLE
› MON_GET_INDEX
› MON_GET_BUFFERPOOL
› MON_GET_TABLESPACE
› MON_GET_CONTAINER
Chapter 4. Monitoring 159
 For example, you see a high Rows Read/Rows Selected ratio and want to
determine the tables that are generating table scans? This is similar to db2pd
–tcbstats but gives you the convenience of relational access to the data. See
Example 4-23.

Example 4-23 Determine the tables that are generating table scan

select tabname, tabschema , table_scans from

table(mon_get_table('','',-2)) as t
order by table_scans desc fetch first 5 rows only"

TABNAME TABLE_SCANS ROWS_READ

---------------------- ------------------- --------------
LAST_TRADE 30419 20789015
CUSTOMER 16149 63507
TRADE_REQUEST 1 11523

You can then query the package cache to see which statements are executed
against these tables and explore the possibility of creating indexes to eliminate
the table scan.

4.6.2 New administrative views

Starting from DB2 9.7 Fixpack 1, there is a new set of administrative views
created based on the new monitoring table functions. They can be used to
access the most important metrics.
› MON_PKG_CACHE_SUMMARY
› MON_CURRENT_SQL
› MON_CURRENT_UOW
› MON_SERVICE_SUBCLASS_SUMMARY
› MON_WORKLOAD_SUMMARY
› MON_CONNECTION_SUMMARY
› MON_DB_SUMMARY
› MON_BP_UTILIZATION
› MON_TBSP_UTILIZATION
› MON_LOCKWAITS

The MONREPORT module provides a set of procedures for retrieving a variety

of monitoring data and generating text reports. Details about the MONREPORT
module is available at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp?topic=/c
om.ibm.db2.luw.sql.rtn.doc/doc/r0056557.html
160 Best Practices for DB2 on AIX 6.1 for POWER Systems
Time-spent metrics is one of the most significant enhancements in monitoring.
We already touched upon it in the MON_GET_ACTIVITY_DETAILS example,
where we used ACT_WAIT_TIME to determine what percentage of total time
SQL spent in wait. Let us take a closer look at it.

Time-spent metrics
Identifying the resource bottleneck is the first and the key step towards any
performance investigation. Understanding the bottleneck can help you rule out a
lot of possible problems early on. See Figure 4-5.

Figure 4-5 Identifying the resource bottleneck

DB2 9.7 introduces the concept of time-spent metrics, a set of metrics that
provides a comprehensive breakdown of where and how time is being spent
inside DB2. They give us valuable clues about the source of problems by telling
where DB2 is spending most of its time. When used with the AIX Monitoring
tools, they can drive the investigation in the right direction. There are three
categories of time-spent metrics:
› Wait Time
› Component Time
› Component Processing Time
Chapter 4. Monitoring 161
Wait Time
This category helps answer the following question: “What percentage of total
time in DB2 is spent on waits, and what are the resources we are waiting on?
Examples are:
› Buffer pool Read/Write Wait Time
(POOL_READ_TIME + POOL_WRITE_TIME)
› Direct Read/Write Wait Time
(DIRECT_READ_TIME + DIRECT_WRITE_TIME)
› Lock Wait Time
(LOCK_WAIT_TIME)
› Log I/O Wait Time
(LOG_DISK_WAIT_TIME + LOG_BUFFER_WAIT_TIME)
› TPC/IP (Send/Receive) Wait Time
(TCPIP_SEND_WAIT_TIME + TCPIP_RECV_WAIT_TIME)
› FCM (Send/Receive) Wait Time
(FCM_SEND_WAIT_TIME + FCM_RECV_WAIT_TIME)

Note: Note that the Buffer pool Read Time and Buffer pool Write time
represent the synchronous read and write time.

Component Time
This category helps answer the following question: “Where am I spending time
within DB2?”

Components in DB2 map to other stages of processing within DB2 (examples,

Sort, Query Compilation, Transaction end processing (Rollback/Commit) and
Utilities). Metrics in this category measure the elapsed time spent across major
DB2 Components. Examples:
› Compile Time
(TOTAL_COMPILE_TIME + TOTAL_IMPLICIT_COMPILE TIME)
› Section Execution Time
(TOTAL_SECTION_TIME)
› Sort Time
(TOTAL_SECTION_SORT_TIME)
162 Best Practices for DB2 on AIX 6.1 for POWER Systems
› Transaction End Processing Time
(TOTAL_COMMIT_TIME + TOTAL_ROLLBACK_TIME)
› Load Processing time
(TOTAL_LOAD_TIME)

Component Processing Time

This category, introduced in DB2 9.7 FixPack1, helps answer the following
question: “How much time is spent doing actual work versus wait within a DB2
Component?” Examples:
› Compile Processing time
(TOTAL_COMPILE_PROC_TIME +
TOTAL_IMPLICIT_COMPILE_PROC_TIME)
› Section Processing time
(TOTAL_SECTION_PROC_TIME)
› Sort Processing time
(TOTAL_SECTION_SORT_PROC_TIME)
Note that Section Processing time includes Sort Processing time
› Transaction End Processing Time
(TOTAL_COMMIT_PROC_TIME + TOTAL_ROLLBACK_PROC_TIME)
› Load Processing time
(TOTAL_LOAD_PROC_TIME)
Chapter 4. Monitoring 163
Using the time spent metrics at other levels, we can get an insight into where we
are spending time in DB2. See Figure 4-6 through Figure 4-8 on page 165.

Figure 4-6 Wait time versus processing time with respect to overall request time

Figure 4-7 Breakup of Total Wait Time

164 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 4-8 Breakup of processing time with component processing time metrics

There is another interesting dimension you can derive from the Time Spent
metrics: you can determine how much time is being spent in the stages of DB2
processing using the Component Time metrics. See Figure 4-9.

Figure 4-9 Other stages of DB2 processing

Chapter 4. Monitoring 165
You can get these perspectives at the system level as well as at the activity level.

Few important points to keep in mind when looking at the time metrics are:
› All time metrics, except for CPU Time, is in milliseconds. CPU Time is in
microseconds.
› All time metrics, including TOTAL_RQST_TIME, is the sum of all work done
by agents working on a query and not the elapsed clock time.
In other words, when we have parallelism between the agents, the values are
greater than elapsed time. TOTAL_APP_RQST_TIME gives the actual
execution time
› Member column correspond to DBPARTITIONNUM of snapshots.
› Sometimes you might find that the CPU Time is less than the processing time
(Total Request time – Total Wait time). This is true in environments where you
have many concurrent connections.
Processing time represents the time an agent is eligible to run. However, it
does not necessarily mean it is consuming CPU. It might be waiting on the
run-queues for the CPU if there are a large number of threads on the system
› MON_FORMAT_XML_TIMES_BY_ROW is a convenient interface to access
the time spent metrics. Visit the following Web page for more information:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp?topic
=/com.ibm.db2.luw.sql.rtn.doc/doc/r0056557.html

Let us see the time-spent metrics in action. A performance problem has been
reported for an OLTP application with ~40 concurrent connections. The
throughput is low at 10 transactions per second. We use the new administrative
views in this example.

What percentage of overall request time in DB2 is spent on processing versus

wait?
› Sum up TOTAL_RQST_TIME and TOTAL_WAIT_TIME using
MON_GET_CONNECTION table function across all active connections.
› Total Processing Time is calculated as TOTAL_RQST_TIME -
TOTAL_WAIT_TIME.
› CLIENT_IDLE_WAIT_TIME is not included in TOTAL_RQST_TIME. A
relatively high client idle wait time might indicate that the problem is not in
DB2.
166 Best Practices for DB2 on AIX 6.1 for POWER Systems
The query converts the time spent metrics as a percentage of total request time,
as shown in Example 4-24.

Example 4-24 Converting time spent metrics as percentage of total request time
WITH TimeSpent AS
( SELECT SUM(TOTAL_WAIT_TIME)AS WAIT_TIME,
SUM(TOTAL_RQST_TIME - TOTAL_WAIT_TIME) AS PROC_TIME,
SUM(TOTAL_RQST_TIME)AS RQST_TIME,
SUM(CLIENT_IDLE_WAIT_TIME) as client_idle_wait_time FROM
TABLE(MON_GET_CONNECTION(NULL,NULL)) AS METRICS)

SELECT
RQST_TIME,
CASE WHEN RQST_TIME > 0
THEN DEC((FLOAT(WAIT_TIME))/FLOAT(RQST_TIME) * 100,5,2)
ELSE NULL END AS WAIT_PCT ,
CASE WHEN RQST_TIME > 0
THEN DEC((FLOAT(PROC_TIME))/FLOAT(RQST_TIME) * 100,5,2)
ELSE NULL END AS PROC_PCT ,
CLIENT_IDLE_WAIT_TIME
FROM TIMESPENT;
RQST_TIME WAIT_PCT PROC_PCT CLIENT_IDLE_WAIT_TIME
-------------------- -------- -------- --------------------- 113200332

40.19 59.80
6959603

It shows that 60% of the total request time is spent in processing while the
remaining 40% is spent in wait.

What is the breakup of the wait times inside DB2? To find this, we query
sysibmadm.mon_db_summary to get summary of the time spent metrics at the
database level. See Example 4-25.

Example 4-25 What is the breakup of the wait times inside DB2

select RQST_WAIT_TIME_PERCENT, IO_WAIT_TIME_PERCENT,

LOCK_WAIT_TIME_PERCENT ,
NETWORK_WAIT_TIME_PERCENT from
sysibmadm.mon_db_summary
RQST_WAIT_TIME_PERCENT IO_WAIT_TIME_PERCENT
 LOCK_WAIT_TIME_PERCENT
NETWORK_WAIT_TIME_PERCENT
---------------------- -------------------- ----------------------
-----------------
38.26 59.76 18.50
5.01

Chapter 4. Monitoring 167

This indicates that 60% of Wait time is spent in I/O Wait. Because OLTP is
characterized by synchronous reads, it is normal to see the majority of wait time
spent in Synchronous Buffer pool Data/Index reads. Is this high enough to be a
concern? To answer this, we need to use AIX monitoring tools to place the
system performance in the right perspective.

In this case, we are running with 40 client threads. Example 4-26 shows how
average CPU use looks in vmstat.

Example 4-26 vmstat

kthr memory page faults cpu
----- ----------- ------------------------ ------------ -----------
r b avm fre re pi po fr sr cy in sy cs us sy id wa 29 9 4544790

1291740 0 0 0 0 0 0 503 3040 3397 97 3 0 0

CPU is saturated at 100% - even if agents are blocked on I/O, we have 40 agents
concurrently executing and they are driving the CPU hard. The Runnable queue
is quite high. So it looks more like a CPU bottleneck. The next obvious question
is where is DB2 spending its time in processing. See Example 4-27.

Example 4-27 Where is DB2 spending its time in processing

select (100 - RQST_WAIT_TIME_PERCENT) PROC_TIME_PERCENT,

SECTION_PROC_TIME_PERCENT, SECTION_SORT_PROC_TIME_PERCENT,
COMPILE_PROC_TIME_PERCENT, TRANSACT_END_PROC_TIME_PERCENT from
sysibmadm.mon_db_summaryl
PROC_TIME_PERCENT SECTION_PROC_TIME_PERCENT
SECTION_SORT_PROC_TIME_PERCENT COMPILE_PROC_TIME_PERCENT
TRANSACT_END_PROC_TIME_PERCENT
----------------- -------------------------
------------------------------ -------------------------
------------------------------
62.37 2.61
0.00 49.83 0.52

1 record(s) selected.

This shows that 50% of our processing time is spent in compiling rather than
executing queries. We are constantly re-compiling queries either because the
package cache is small or possibly because we are using the same SQL with
other literals. We can further check the package cache hit ratio and monitor the
 SQLs to see if there is a possibility to use parameter markers, instead of literals.

The time-spent metrics, along with vmstat, helped us identify the source of the
performance issue.

168 Best Practices for DB2 on AIX 6.1 for POWER Systems
 The time-spent metrics are available at both the system level and activity level.
They are also available in the new event monitors introduced in DB2 9.7. More
details about the Time Spent Metrics and other monitoring enhancements are
available in DB2 9.7 Information Center:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp?topic=/c
om.ibm.db2.luw.wn.doc/doc/c0056333.html

4.7 Monitoring tools for AIX

AIX provides a lot of utilities that can help analyze or identify a performance
bottleneck when a DB2 application is deployed and while serving several
workloads. This section describes these utilities along with useful tools that can
be used to identify the bottlenecks. The following performance thresholds were
observed in these tools or common AIX utilities. This can be used as reference
point to identify performance bottlenecks.
› CPU bound
When %user + %sys is greater than 80% (Tool used is vmstat).
› Disk I/O bound
When %iowait is greater than 40% (Common utility used is vmstat).
› Application disk bound
When %tm_act is greater than 70% (common utility used is iostat).
› Paging space low
When % paging space used is greater than 70% active (common utility is lsps
-a).
› Paging bound
When paging logical volumes %tm_act is greater than 30% of the I/O
(common utility is iostat) and paging activity is greater than 10 times the
number of CPUs (common utility is vmstat).
› Thrashing
Rising page outs, CPU wait, and run queue high (common utility is vmstat).
Chapter 4. Monitoring 169
 Tip: Although most of the utilities can be used in various ways to identify
performance bottlenecks, nmon can be used as a single or general tool that
can provide all of most of the data on a single platform to achieve the same
objective.

Details on installing and using nmon can be found in the following Web page:
http://www.ibm.com/developerworks/aix/library/au-analyze_aix/

We have a separate section dedicated to nmon to monitor AIX (see “nmon” on

page 179) and how it maps to the existing AIX utilities that is used traditionally
by performance engineers.

In this section, we focus on using these utilities and tools to help identify the
bottlenecks at various Db2 workloads while monitoring DB2.
Virtual memory management statistics, VMSTAT
The vmstat command is used to report statistics about kernel threads in the run
and wait queues, memory, paging, disks, interrupts, system calls, context
switches, and CPU activity. If the vmstat command is used without any options or
only with the interval and optionally, the count parameter, such as vmstat 2, then
the first line of numbers is an average since system reboot.
Syntax
vmstat interval count
Reporting
The column headings in the report are:
› r: Number of processes on run queue per second
› b: Number of processes awaiting paging in per second
› avm: Active virtual memory pages in paging space
› fre: Real memory pages on the free list
› re: Page reclaims; free, but claimed before being reused
› pi: Paged in (per second)
› po: Paged out (per second)
› fr: Pages freed (page replacement per second)
› sr: Pages per second scanned for replacement
› cy: Complete scans of page table
› in: Device interrupts per second
› sy: System calls per second
› cs: CPU context switches per second
› us: User CPU time percentage
› sys: System CPU time percentage
› id: CPU idle percentage (nothing to do)
› wa: CPU waiting for pending local Disk I/O
170 Best Practices for DB2 on AIX 6.1 for POWER Systems
The important columns in VMSTAT output to look for are id, sys, wa, and us while
monitoring CPU of the AIX system.

Figure 4-10 shows a case of a high CPU issue being identified using the
VMSTAT utility. The system is hitting an average of 65% user CPU usage (us
column) and 35% system (sy column) CPU usage. Pi and Po column values are
equal to 0, so there are no paging issues. The wa column shows there does not
seem to be any I/O issues. See Figure 4-10.

Figure 4-10 vmstat output

The way to interpret VMSTAT for CPU usage in a virtualization environment is a

bit different from the traditional way. The CPU usage in a LPAR is dependent on
the type of partitioning used and type of mode. For details on this, see the
following Web page:
http://www.ibm.com/developerworks/wikis/display/WikiPtype/Virtualizatio
n+Concepts

The VMSTAT snapshot in Example 4-28 on page 172 shows the system
monitoring activity on an uncapped micro portioning environment with initial CPU
entitlement as 0.4, highlighted as “ENT” in the output. With shared pools and
LPAR micro partitioning being enabled, note two additional columns as PC and
EC shown in VMSTAT output.

PC represents the physical consumed processors and EC represents

percentage of entitlement consumed. Example 4-28 on page 172 shows the
initial entitlement is 0.4 CPU with physical consumed up to 2.57 CPU being used.
In other words, the user and system cpu usage is relative to physical consumed
and initial entitlement values.

EC represents entitled capacity. EC represents entitled capacity. This value is

related to the initial CPU entitlement. Depending on the initial CPU entitlement,
the entitled capacity shows how much % of the initial CPU entitlement is being
used.
Chapter 4. Monitoring 171
 In Example 4-28, with uncapped partition, the EC value is more than 100 as the
initial CPU entitlement is 0.4 CPU and when the physical consumed, PC is more
than 0.4 CPU, EC goes more than 100.

Example 4-28 VMSTAT snapshot

System configuration: lcpu=8 mem=16384MB ent=0.40

kthr memory page faults cpu

----- ----------- ------------------------ ------------ -----------------------
r b avm fre re pi po fr sr cy in sy cs us sy id wa pc ec
5 0 2493379 965450 0 0 0 0 0 0 426 20400 5727 86 12 2 0 0.66 165.8
3 0 2493383 965328 0 0 0 0 0 0 541 25948 7791 96 4 1 0 2.57 643.1
5 0 2493414 965219 0 0 0 0 0 0 486 24601 7551 95 4 1 0 2.17 542.9
3 0 2493410 965166 0 0 0 0 0 0 410 21398 6209 92 7 1 0 1.25 313.0
3 0 2493421 965115 0 0 0 0 0 0 376 19853 6593 95 4 1 0 1.79 446.5
3 0 2493452 964993 0 0 0 0 0 0 547 25833 7323 94 5 1 0 1.91 476.5
4 0 2493454 964918 0 0 0 0 0 0 466 26966 7848 95 4 1 0 2.12 530.2
6 0 2493472 964809 0 0 0 0 0 0 554 30049 8407 94 5 1 0 2.22 553.9

Tip: VMSTAT can be used to identify memory bottlenecks. If the pi and po

values are not zero, it is most likely a memory issue. Occasional values of non
zero are OK. For details refer to the following Web page:
http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.
aix.prftungd/doc/prftungd/mem_perf.htm

The snapshot from VMSTAT in Figure 4-11 shows the wa (waiting on I/O) column
to be unusually high. This indicates there might be I/O bottlenecks on the system,
which in turn, causes the CPU usage to be inefficient.

Figure 4-11 VMSTAT snapshot

172 Best Practices for DB2 on AIX 6.1 for POWER Systems
Disk IO statistics: IOSTAT
The iostat command is used to report CPU and I/O statistics for TTY devices,
disks, and CD-ROMs. It is used to generate reports that can be used to change
the system configuration to better balance the input/output load between physical
disks.

Syntax
iostat interval count

Reporting columns
Reporting columns are:
› %tm_act
Reports back the percentage of time that the physical disk was active or the
total time of disk requests.
› Kbps
Reports back the amount of data transferred to the drive in kilobytes.
› tps
Reports back the number of transfers-per-second issued to the physical disk.
› Kb_read
Reports back the total data (kilobytes) from your measured interval that is
read from the physical volumes.
› Kb_wrtn
Reports back the amount of data (kilobytes) from your measured interval that
is written to the physical volumes.

See Figure 4-12.

Figure 4-12 Snaphot from IOSTAT showing greater value than 40% to report a IO
bottleneck
Chapter 4. Monitoring 173
To check if there is resource contention, focus on the %tm_act value from the
output. An increase in this value, especially more than 40%, implies that
processes are waiting for I/O to complete, and there is an I/O issue. For more
details on IOSTAT, visit the AIX Information Center:
http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/
com.ibm.aix.prftungd/doc/prftungd/iostat_command.htm

In case of an AIX partitioning environment, the IOSTAT header changes. The

portion of IOSTAT header output in Figure 4-13 shows an uncapped partitioning
LPAR with initial CPU entitlement of 0.40. Remember that physc % is physical
consumed and is similar to the VMSTAT PC column discussed in the VMSTAT
section.

Figure 4-13 IOSTAT header output

List paging space: lsps

The lsps command displays the characteristics of paging spaces, such as the
paging space name, physical volume name, volume group name, size,
percentage of the paging space used, and whether the space is active or
inactive. See Example 4-29.

Example 4-29 List paging space

lsps -a
Page Space Physical Volume Volume Group Size %Used Active
Auto Type
paging01 hdisk0 rootvg 1024MB 3 yes
yes lv
paging00 hdisk0 rootvg 1024MB 3 yes
yes lv
hd6 hdisk0 rootvg 512MB 0

no yes lv

Logical volume manager statistic: lvmstat

The lvmstat command helps you identify and remedy hot-spot logical partitions
(LPs) within your logical volumes. The gathering of statistics has to be enabled
first with the lvmstat command for either a logical volume or an entire volume
group. When identified, use migratelp to move those hot-spot LPs elsewhere.
174 Best Practices for DB2 on AIX 6.1 for POWER Systems
The following commonly used flags are for the lvmstat command:
› –e
Enables the gathering of statistics about the logical volume.
› -d
Disables the gathering of statistics
› -l
Identifies the target logical volume
› -v
Specifies the target volume group
› -F
Specifies colon separated output

See the following Web page for details on using LVMSTAT:

http://publib.boulder.ibm.com/infocenter/aix/v6r1/index.jsp?topic=/com.
ibm.aix.cmds/doc/aixcmds3/lvmstat.htm

Process State PS
This utility is used to display the status of currently active processes. Apart from
using this utility to understand the CPU of a active process, this utility can be
used to diagnose or identify performance issues (such as memory leak and hang
situations). For details on PS utility usage, see the following Information Center
Web page:
http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/
com.ibm.aix.prftungd/doc/prftungd/ps_command.htm
Chapter 4. Monitoring 175
TOPAS
This utility is used to get the CPU usage of a certain process in the AIX system. It
is used to identify which the processes are consuming most CPU in the AIX
system when VMSTAT command whows that there is a CPU bottleneck in the
system. See Figure 4-14.

Figure 4-14 Snapshot from topas is used to establish that the DB2 engine is contributing
to the CPU bottleneck

topas can be used with several options. One common option used in virtual
environment is topas –C. This command collects a set of metrics from AIX
partitions running on the same hardware platform. For details on using TOPAS,
see the following Web page:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm.aix.cmd
s/doc/aixcmds5/topas.htm

Netstat
The netstat command is used to show network status. It can be used to
determine the amount of traffic in the network to ascertain whether performance
problems are due to network congestion. See the following Web page for details
on its usage:
http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix
.prftungd/doc/prftungd/using_netstat.htm

System Virtual Memory Monitor: SVMON

The svmon command is used to capture and analyze a snapshot of virtual
memory.
176 Best Practices for DB2 on AIX 6.1 for POWER Systems
Report
The following are the column headings of the report:
› size
In pages (4096)
› inuse
In use
› free
Not in use (including rmss pages)
› pin
Pinned (locked by application)
› work
Pages in working segments
› pers
Pages in persistent segments
› clnt
Pages in client segments
› pg
space Paging space

SVMON –G is a common utility that is used to present global statistics and can help
in getting the available memory in system. This is quite effective in finding real
memory estimate used by DB2 instance.

Tip: The VMSTAT –v command can be used to get the similar function.
Chapter 4. Monitoring 177
SVMON can be used to monitor the number of 4-kilobyte and 64-kilobyte page
frames on a AIX Power system. For example, to display DB2 process statistics
about each page size, use the -P flag with the DB2 process ID (PID) with the
svmon command. Figure 4-15 shows the same.

Figure 4-15 SVMON

178 Best Practices for DB2 on AIX 6.1 for POWER Systems
 nmon
nmon is an AIX monitoring tool that can used to capture and monitor most of the
system related data. The beauty of using nmon is by using only a single utility,
several system monitoring parameters relevant from DB2 point of view can be
analyzed. nmon makes the analysis with more ease by generating a spreadsheet
automatically, with the help of nmon analyzer.

Figure 4-16 shows the first panel seen when nmon is used. The first panel shows
the AIX level and Processor type details. Use nmon –h for full details of usage
help.

Figure 4-16 AIX and processor details

Chapter 4. Monitoring 179
 The nmon can be used to display all the AIX system monitoring data on one panel
and it can get updated periodically. When running, type h for further help on the
options or type q to quit. Figure 4-17 and Figure 4-18 on page 181 show most of
the similar system parameters such as CPU, Memory, Paging, Disk, Network and
so forth, which we monitored by using several AIX utilities relevant to monitor
DB2 application.

Figure 4-17 nmon output

180 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 4-18 nmon output

There are separate sections in one panel that represent separate system
monitoring parameter. Each of these can be mapped to its corresponding AIX
monitoring utility as we discussed in the earlier sections in this chapter. For
example, the Topas command equivalent can be seen in “Top Processes” section
of nmon. Similarly, VMSTAT o/p equivalent can be seen in
“CPU-Utilization-Small-View” section of the nmon panel.

The performance threshold also applies to this and can be used as a reference to
identify using nmon if a performance bottleneck is encountered or not. For
example, if there is a CPU bound situation, check the
"CPU-Utilization-Small-View“ section of the nmon output as shown in Figure 4-19
on page 182. If the user% + sys% goes beyond 80%, the system is CPU bound.
Chapter 4. Monitoring 181
 Figure 4-19 shows in this case, the user% is 77% and sys% is 2.2%, which is
about 80%. The Wait% is zero, signifying that there is CPU waiting for other
activity (such as IO) to be completed.

Figure 4-19 nmon output

nmon is much simpler to use compared to remembering so many utilities for

monitoring AIX. On top of this, nmon analyzer helps to produce a spread sheet
automatically, which makes it easier to see the system vitals in the form of charts
and graphs. Refer to the following Web page learn more about nmon analyzer:
http://www.ibm.com/developerworks/aix/library/au-nmon_analyser/

Figure 4-20 provides a sample CPU usage pattern captured over a period of time
by nmon and generated from nmon analyzer. This is equivalent for VMSTAT
output shown in Figure 4-19.

Figure 4-20 Sample CPU usage pattern

182 Best Practices for DB2 on AIX 6.1 for POWER Systems
 These snapshots show that there is good amount of system% activity with % wait
time by the CPU’s. Figure 4-21 shows the status of the active processes.

Figure 4-21 Status of the active processes

Chapter 4. Monitoring 183
 Figure 4-22 shows a network usage chart captured by nmon and shown by the
nmon analyzer. The figure shows that there is a network issue that was getting
worse under high load situations. The pattern highlighted in Figure 4-22 shows
that there is a network issue due to which there is no network IO activity in a few
occasions. This is similar to “netstat” and “tcpdump” analysis data that is required
to monitor the network traffic.

Figure 4-22 Network usage chart

In a virtualization environment, the nmon output is a bit different to the traditional

nmon output. Figure 4-23 shows LPAR-related information as shown by nmon,
which highlights the initial CPU entitlement value as compared to the virtual
CPUs.

Figure 4-23 nmon graph

184 Best Practices for DB2 on AIX 6.1 for POWER Systems
 The nmon graph in Figure 4-24 shows the CPU behavior of an uncapped
partition LPAR with an initial entitlement of 1.2 CPU’s. Thus, CPU behavior is
again relative to the entitled capacity and the physical processor’s that are being
consumed.

Figure 4-24 nmon graph

4.8 Monitoring scenarios

In this section we cover several real life monitoring scenarios.

4.8.1 Case 1: Alter Tablespace no file system cache

In this scenario, we have applications connecting to DB2 through WebSphere
Application Server. These applications, which perform inserts and updates, are
experiencing a severe performance degradation. The applications have been
ported onto the current system from another where the data was identical to the
original environment in which the application performed as expected. Given this
situation, one can employ a top-down approach to determine the cause of this
degradation.
Chapter 4. Monitoring 185
Let us assume that it has been determined that the problem does not lie in
WebSphere Application Server. At this point we start looking into the DB2 and
AIX monitoring information. See Example 4-30.

Example 4-30 vmstat output

vmstat
kthr memory page faults cpu
----- ----------- ------------------------ ------------ -----------
r b avm fre re pi po fr sr cy in sy cs us sy id wa
2 2 5187657 10998 0 0 0 863 963 0 1693 79115 7175 3 12 67
18
4 1 5187621 10914 0 0 0 167 202 0 571 4751 4892 0 26 54 19
1 1 5187559 10532 0 0 0 0 0 0 1433 19293 8553 2 9 73
16
20 0 5187614 10230 0 0 0 0 0 0 656 6603 4921 1 14 63 22
5 0 5187657 10058 0 0 0 128 146 0 873 6912 5784 1 34 53 12
8 0 5187656 10107 0 0 0 138 145 0 230 3707 3630 1 25 63 11
0 0 5187669 10008 0 0 0 1 0 0 403 10491 4611 1 25 60 14
8 0 5187685 9972 0 0 0 59 73 0 573 7408 5146 1 23 62 14
11 2 5187833 9915 0 0 0 483 540 0 274 38571 3721 3 46 36 15
7 2 5187709 10598 0 0 0 420 484 0 222 3827 3691 1 34 50 15

Starting with the vmstat output it is not apparent as to where exactly the
bottleneck can be, even though we see plenty of kernel threads on the run queue
and likewise with time spent in wait. The focus now turns to DB2 data, which is
collected when the performance degradation was observed. Again, we employ a
top-down approach starting with the dbm snapshot and working down to
individual queries if needed. This is where the quality of information becomes so
important. Based on the quality of this information, one can narrow down and
hypothesize about where the problem can lie. In our case a 3x degradation is
experienced by insert and update statements for all the applications. Knowing
this, we can look into factors that affect more than a query, as opposed to say an
access plan for any given query. See Example 4-31.

Example 4-31 Initial Database Manager Snapshot

Database Manager Snapshot

Node type = Enterprise Server

Edition with local and remote clients
Instance name = DB2INST1
Number of database partitions in DB2 instance =1
Database manager status = Active
 Private Sort heap allocated =0

186 Best Practices for DB2 on AIX 6.1 for POWER Systems
Private Sort heap high water mark = 7050
Post threshold sorts = 0
Piped sorts requested = 3644324
Piped sorts accepted = 3644324

Start Database Manager timestamp = 11/03/2009

14:29:58.578177
Last reset timestamp =
Snapshot timestamp = 11/06/2009
20:59:51.493364
Remote connections to db manager = 63
Remote connections executing in db manager = 5
Local connections = 1
Local connections executing in db manager = 0
Active local databases = 1

High water mark for agents registered = 93

High water mark for agents waiting for a token = 0
Agents registered = 93
Agents waiting for a token = 0
Idle agents = 23

Committed private Memory (Bytes) = 1916928

Switch list for db partition number 0

Buffer Pool Activity Information = ON 11/03/2009
(BUFFERPOOL
)
14:29:58.578177
Lock Information (LOCK) = ON 11/03/2009
14:29:58.578177
Sorting Information (SORT) = ON 11/03/2009
14:29:58.578177
SQL Statement Information = ON 11/03/2009
(STATEMENT
)
14:29:58.578177
Table Activity Information (TABLE) = ON 11/03/2009
14:29:58.578177
Take Timestamp Information = ON 11/03/2009
(TIMESTAMP
)
14:29:58.578177
Unit of Work Information = ON 11/03/2009
(UOW
)

14:29:58.578177

Agents assigned from pool = 5408

Agents created from empty pool = 115
Agents stolen from another application = 0
High water mark for coordinating agents = 93

Chapter 4. Monitoring 187

Max agents overflow =0
Hash joins after heap threshold exceeded =0

Total number of gateway connections =0

Current number of gateway connections = 0 Gateway
connections waiting for host reply = 0 Gateway
connections waiting for client request = 0 Gateway

connection pool agents stolen =0

Again, we cannot deduce where the problem lies so we proceed with the next set
of snapshots, the database snapshots, which are taken 70 seconds apart.
Example 4-32 shows the first snapshot.

Example 4-32 Database Snapshot taken at 11/06/2009 20:59:51.556288

Database Snapshot

Database name = SAMPLE

Database path =
/home/db2inst1/db2inst1/NODE0000/SQL00001/
Input database alias = SAMPLE
Database status = Active
Catalog database partition number =0
Catalog network node name =
Operating system running at database server= AIX
Location of the database = Local
First database connect timestamp = 11/03/2009 14:31:55.028855
Last reset timestamp =
Last backup timestamp = 11/06/2009 04:26:58.000000
Snapshot timestamp = 11/06/2009 20:59:51.556288

Number of automatic storage paths =0

High water mark for connections = 91

Application connects = 3383
Secondary connects total =0
Applications connected currently = 64
Appls. executing in db manager currently =7
Agents associated with applications = 64
Maximum agents associated with applications= 91
Maximum coordinating agents = 91
 Locks held currently = 202
Lock waits = 5343
Time database waited on locks (ms) = 2019035

188 Best Practices for DB2 on AIX 6.1 for POWER Systems
Lock list memory in use (Bytes) = 62960
Deadlocks detected = 10
Lock escalations = 198
Exclusive lock escalations = 177
Agents currently waiting on locks = 0
Lock Timeouts = 52
Number of indoubt transactions = 0

Total Private Sort heap allocated = 0

Total Shared Sort heap allocated = 0
Shared Sort heap high water mark = 0
Total sorts = 3644404
Total sort time (ms) = 3000518
Sort overflows = 2089
Active sorts = 0

Buffer pool data logical reads = 573598824

Buffer pool data physical reads = 100130713
Buffer pool temporary data logical reads = 268442132
Buffer pool temporary data physical reads = 6240450
Asynchronous pool data page reads = 91774524
Buffer pool data writes = 15181373
Asynchronous pool data page writes = 14919461
Buffer pool index logical reads = 913437418
Buffer pool index physical reads = 9632933
Buffer pool temporary index logical reads = 0
Buffer pool temporary index physical reads = 0
Asynchronous pool index page reads = 1966562
Buffer pool index writes = 4859808
Asynchronous pool index page writes = 4838753
Total buffer pool read time (milliseconds) = 80594078
Total buffer pool write time (milliseconds)= 16800361
Total elapsed asynchronous read time = 20606257
Total elapsed asynchronous write time = 15088127
Asynchronous data read requests = 4176336
Asynchronous index read requests = 133529
No victim buffers available = 1817877
LSN Gap cleaner triggers = 0
Dirty page steal cleaner triggers = 0
Dirty page threshold cleaner triggers = 0
Time waited for prefetch (ms) = 12712476
Unread prefetch pages = 99918
Direct reads = 295290549
0
Direct writes = 20900113
Direct read requests = 37386554
Chapter 4. Monitoring 189
Direct write requests = 1342561
Direct reads elapsed time (ms) = 96245210
Direct write elapsed time (ms) = 28184965
Database files closed = 0
Data pages copied to extended storage = 0
Index pages copied to extended storage = 0
Data pages copied from extended storage = 0
Index pages copied from extended storage = 0

Host execution elapsed time = 2.346626

Commit statements attempted = 8665942

Rollback statements attempted = 645683
Dynamic statements attempted = 71548695
Static statements attempted = 5686790
Failed statement operations = 176729
Select SQL statements executed = 20914657
Update/Insert/Delete statements executed = 13160495
DDL statements executed = 110
Inactive stmt history memory usage (bytes) = 0

Internal automatic rebinds = 0

Internal rows deleted = 8588236
Internal rows inserted = 1254435
Internal rows updated = 0
Internal commits = 3383
Internal rollbacks = 123
Internal rollbacks due to deadlock = 71

Rows deleted = 30132067

Rows inserted = 30161713
Rows updated = 2627441
Rows selected = 61559544
Rows read = 9500034756
Binds/precompiles attempted = 0

Log space available to the database (Bytes)= 9928132139

Log space used by the database (Bytes) = 98875861
Maximum secondary log space used (Bytes) = 0
Maximum total log space used (Bytes) = 2927224730
Secondary logs allocated currently = 0
Log pages read = 461948
Log read time (sec.ns) = 22.000000004
Log pages written = 9057242
Log write time (sec.ns) = 8867.000000004
190 Best Practices for DB2 on AIX 6.1 for POWER Systems
Number write log IOs = 3779013
Number read log IOs = 0
Number partial page log IOs = 1817223
Number log buffer full = 420
Log data found in buffer = 8221005
Appl id holding the oldest transaction = 1268
Log to be redone for recovery (Bytes) = 45237972
Log accounted for by dirty pages (Bytes) = 45237972

File number of first active log = 77623

File number of last active log = 77657
File number of current active log = 77623
File number of log being archived = Not applicable

Package cache lookups = 25999275

Package cache inserts = 683862
Package cache overflows = 0
Package cache high water mark (Bytes) = 4449189
Application section lookups = 76454994
Application section inserts = 822653

Catalog cache lookups = 5079469

Catalog cache inserts = 7738
Catalog cache overflows = 0
Catalog cache high water mark = 0

Workspace Information

Shared high water mark = 0

Corresponding shared overflows = 0
Total shared section inserts = 0
Total shared section lookups = 0
Private high water mark = 5332000
Corresponding private overflows = 0
Total private section inserts = 822653
Total private section lookups = 23453369

Number of hash joins = 31086

Number of hash loops = 22
Number of hash join overflows = 47
Number of small hash join overflows = 1
Chapter 4. Monitoring 191
Example 4-33 shows the second snapshot, which was taken 70 seconds after
the first snapshot.

Example 4-33 Subsequent database snapshot (11/06/2009 21:01:00.612305)

Database Snapshot

Database name = SAMPLE

Database path =
/db2inst1/db2inst1/NODE0000/SQL00001/
Input database alias = SAMPLE
Database status = Active
Catalog database partition number =0
Catalog network node name =
Operating system running at database server= AIX
Location of the database = Local
First database connect timestamp = 11/03/2009 14:31:55.028855
Last reset timestamp =
Last backup timestamp = 11/06/2009 04:26:58.000000
Snapshot timestamp = 11/06/2009 21:01:00.612305

Number of automatic storage paths =0

High water mark for connections = 91

Application connects = 3384
Secondary connects total =0
Applications connected currently = 64
Appls. executing in db manager currently =3
Agents associated with applications = 64
Maximum agents associated with applications= 91
Maximum coordinating agents = 91

Locks held currently = 205

Lock waits = 5345
Time database waited on locks (ms) = 2021212
Lock list memory in use (Bytes) = 64960
Deadlocks detected = 10
Lock escalations = 198
Exclusive lock escalations = 177
Agents currently waiting on locks =0
Lock Timeouts = 52
Number of indoubt transactions =7
 Total Private Sort heap allocated =0
Total Shared Sort heap allocated =0

192 Best Practices for DB2 on AIX 6.1 for POWER Systems
Shared Sort heap high water mark = 0
Total sorts = 3645714
Total sort time (ms) = 3000542
Sort overflows = 2089
Active sorts = 0

Buffer pool data logical reads = 573845323

Buffer pool data physical reads = 100132787
Buffer pool temporary data logical reads = 268442849
Buffer pool temporary data physical reads = 6240450
Asynchronous pool data page reads = 91774540
Buffer pool data writes = 15181663
Asynchronous pool data page writes = 14919751
Buffer pool index logical reads = 913523930
Buffer pool index physical reads = 9636215
Buffer pool temporary index logical reads = 0
Buffer pool temporary index physical reads = 0
Asynchronous pool index page reads = 1966562
Buffer pool index writes = 4861015
Asynchronous pool index page writes = 4839960
Total buffer pool read time (milliseconds) = 80751657
Total buffer pool write time (milliseconds)= 16805469
Total elapsed asynchronous read time = 20607017
Total elapsed asynchronous write time = 15093235
Asynchronous data read requests = 4176352
Asynchronous index read requests = 133529
No victim buffers available = 1817877
LSN Gap cleaner triggers = 0
Dirty page steal cleaner triggers = 0
Dirty page threshold cleaner triggers = 0
Time waited for prefetch (ms) = 12713235
Unread prefetch pages = 99918
Direct reads = 2952931034
Direct writes = 20904061
Direct read requests = 37395065
Direct write requests = 1342873
Direct reads elapsed time (ms) = 96329593
Direct write elapsed time (ms) = 28385885
Database files closed = 0
Data pages copied to extended storage = 0
Index pages copied to extended storage = 0
Data pages copied from extended storage = 0
Index pages copied from extended storage = 0

Host execution elapsed time = 2.344887

Chapter 4. Monitoring 193
Commit statements attempted = 8668913
Rollback statements attempted = 645683
Dynamic statements attempted = 71567629
Static statements attempted = 5687093
Failed statement operations = 176797
Select SQL statements executed = 20919560
Update/Insert/Delete statements executed = 13162309
DDL statements executed = 110
Inactive stmt history memory usage (bytes) = 0

Internal automatic rebinds = 0

Internal rows deleted = 8588250
Internal rows inserted = 1254436
Internal rows updated = 0
Internal commits = 3384
Internal rollbacks = 123
Internal rollbacks due to deadlock = 71

Rows deleted = 30132088

Rows inserted = 30163209
Rows updated = 2627637
Rows selected = 61563275
Rows read = 9502045889
Binds/precompiles attempted = 0

Log space available to the database (Bytes)= 9926403251

Log space used by the database (Bytes) = 100604749
Maximum secondary log space used (Bytes) = 0
Maximum total log space used (Bytes) = 2927224730
Secondary logs allocated currently = 0
Log pages read = 461948
Log read time (sec.ns) = 22.000000004
Log pages written = 9058007
Log write time (sec.ns) = 8886.000000004
Number write log IOs = 3779594
Number read log IOs = 0
Number partial page log IOs = 1817491
Number log buffer full = 420
Log data found in buffer = 8221005
Log to be redone for recovery (Bytes) = 46550039
Log accounted for by dirty pages (Bytes) = 46550039

File number of first active log = 77623

File number of last active log = 77657
194 Best Practices for DB2 on AIX 6.1 for POWER Systems
File number of current active log = 77623
File number of log being archived = Not applicable

Package cache lookups = 26005109

Package cache inserts = 683897
Package cache overflows = 0
Package cache high water mark (Bytes) = 4449189
Application section lookups = 76474230
Application section inserts = 822754

Catalog cache lookups = 5079477

Catalog cache inserts = 7739
Catalog cache overflows = 0
Catalog cache high water mark = 0

Workspace Information

Shared high water mark = 0

Corresponding shared overflows = 0
Total shared section inserts = 0
Total shared section lookups = 0
Private high water mark = 5332000
Corresponding private overflows = 0
Total private section inserts = 822754
Total private section lookups = 23459051

Number of hash joins = 31131

Number of hash loops = 22
Number of hash join overflows = 47
Number of small hash join overflows = 1

Observations based on these snapshots

Here are our observations:
› Locking does not seem to be an issue as the Time database waited on locks
(ms) = 2 seconds.
› There does not appear to be a problem with the sorting either as the
difference in 'Total sort time (ms)' is negligible.
› The data page hit ratio is 82%. This is somewhat borderline however, it
cannot cause the 3x performance degradation.
› The Index page hit ratio is 98% which is good.
Chapter 4. Monitoring 195
› The amount of time spent in the directly reading and writing pages is quite
high though, as shown:
Direct reads elapsed time (ms) = 96245210
Direct write elapsed time (ms) = 28184965
Direct reads elapsed time (ms) = 96329593
Direct write elapsed time (ms) = 28385885
Direct read = 84383msec
84 seconds
Direct write = 200920msec
200 seconds

This leads us to the suspicion that the bottleneck might be in I/O. Based on this
we need to determine which table spaces this I/O is performed against. To find
this information we observe the table space snapshot and look for a table space
that shows high direct reads and writes as per the previous findings. This reveals
that LONG_TS is the table space against which the vast majority of direct reads
and writes are spending the most time. See Example 4-34.

Example 4-34 Tablespace snapshot

Tablespace name = TBSP1
Tablespace ID = 1
Tablespace Type = Database managed space
Tablespace Content Type = Long data only
Tablespace Page size (bytes) = 4096
Tablespace Extent size (pages) = 32
Automatic Prefetch size enabled = No
Tablespace Prefetch size (pages) = 32
Buffer pool ID currently in use = 1
Buffer pool ID next startup = 1
Using automatic storage = No
Auto-resize enabled = No
File system caching = Yes
Tablespace State = 0x'00000000'
Detailed explanation: Normal
Total number of pages = 27848630
Number of usable pages = 27848576
Number of used pages = 23445856
Number of pending free pages = 0
Number of free pages = 4402720
High water mark (pages) = 27407776
Current tablespace size (bytes) = 2398838784
Rebalancer Mode = No Rebalancing
Minimum Recovery Time = 06/07/2003 03:58:05.000000
 Number of quiescers = 0

196 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Number of containers =1

Container Name =
/db2/SAMPLE/TBSP1 Container ID =0
Container Type = File (extent sized tag)
Total Pages in Container = 27848630
Usable Pages in Container = 27848576
Stripe Set =0
Container is accessible = Yes

Table space map:

Range Stripe Stripe Max Max Start End Adj.

Containers
Number Set Offset Extent Page Stripe Stripe
[ 0] [ 0] 0 870267 27848575 0 870267 0 1 (0)

Buffer pool data logical reads = 17032043

Buffer pool data physical reads = 14343
Buffer pool temporary data logical reads =0
Buffer pool temporary data physical reads =0
Asynchronous pool data page reads = 0
Buffer pool data writes = 10082
Asynchronous pool data page writes = 10053
Buffer pool index logical reads = 0
Buffer pool index physical reads = 0
Buffer pool temporary index logical reads =0
Buffer pool temporary index physical reads =0
Asynchronous pool index page reads = 0
Buffer pool index writes = 0
Asynchronous pool index page writes = 0
Total buffer pool read time (millisec) = 300846
Total buffer pool write time (millisec) = 14893
Total elapsed asynchronous read time = 0
Total elapsed asynchronous write time = 13715
Asynchronous data read requests = 0
Asynchronous index read requests = 0
No victim buffers available = 34
Direct reads = 636260731
Direct writes = 10830778
Direct read requests = 1669858
Direct write requests = 1219477
Direct reads elapsed time (ms) = 49280622
Direct write elapsed time (ms) = 26318311
Number of files closed = 0

Chapter 4. Monitoring 197

 Data pages copied to extended storage = 0 Index
pages copied to extended storage = 0 Data pages
copied from extended storage = 0 Index pages

copied from extended storage = 0

So, what is contained in this table space? Looking at the Tablespace Content
Type we can see that it is Long Data Only. This gives us a good piece of
information that the problem has something to do with the reading and writing of
long data (for example, clobs/lobs) against this table space. However, consider
the output of other monitoring data before reaching any conclusion.

At this point, rather than proceeding with looking into the buffer pools/tables and
various other snapshot data, let us focus our attention to tracing from a DB2 and
AIX point of view. Using the perfcount option of a DB2 trace, it can be seen in
Example 4-35 that DB2 is spending a significant amount of time in the following
DB2 functions:

Example 4-35 Perfcount option of a DB2 trace

EXECUTIONS SECONDS C3
----------- ----------- -----------------------
112 263 sqldCompleteAsyncWrite
743 254 sqldxWriteLobData
757 254 sqldx_diskwrite
743 254 sqldxReplaceLob
721 253 sqldxCreateLob
721 253 sqldxLobCreate

DB2 trace: The db2trc command controls the trace facility of a DB2 instance
or the DB2 Administration Server (DAS). The trace facility records information
about operations and puts this information into a readable format.

Even without deep knowledge of DB2, you can estimate that functions
sqldxWriteLobData, sqldxReplaceLob, sqldxCreateLob, and sqldxLobCreate
appear to be functions that manipulates lob data on disk. This is consistent with
the observation regarding direct writes and reads against a long table space.

At this point we know that DB2 is spending a considerable amount of time

manipulating lobs against this table space. Is this abnormal or expected if there is
a high amount of activity that performs this sort of work? Again, the quality of
data comes into play. It is determined that the application does process a
significant amount of lob data.
198 Best Practices for DB2 on AIX 6.1 for POWER Systems
 We know there is something suspicious about the I/O against this table space.
Now let us go back to the definition of this table space to see if there is something
that perhaps is changed or has been overlooked. From the table space snapshot
we can see that this table space uses file system caching.
File system caching = Yes

Even though I/O operations against a DMS Long Data table spaces are direct
I/O, we might not be taking full advantage of concurrent I/O operations against
this table space. At this point it was decided to alter the table space to disallow
file system caching, which permits concurrent I/O against the table space
containers.

db2 alter tablespace TBSP1 NO FILE SYSTEM CACHING

Having made this change, the application throughput increased back to the
expected level, that is, the 3x performance degradation was alleviated.

4.8.2 Case 2: Alleviating bottlenecks during index creation

This scenario demonstrates what happens as a result of not configuring the
various DB2 configuration parameters and table space attributes correctly. This
leads to excessive I/O and, coupled with the serialization of the tasks involved,
drastically affects the impact of the SQL statement. The scenario is the creation
of a simple index. However, as it is shown, by taking simple precautions and
monitoring exactly what is happening, we can increase the speed of an index
creation by up to four times. Let us use the table space definitions in
Example 4-36 and Example 4-37 on page 200 as our example.

Example 4-36 Tablespace definitions

[663] db2ins29@eva88 /home2/db2ins29/pmr/b> cat crttbsp.sql

create large tablespace datadms

pagesize 8192
managed by database
using ( file '/pmr/amar/dataconta' 40000 , file
'/pmr/amar/datacontb' 40000 ,
file '/home2/db2ins29/pmr/b/datacontc' 40000 ,
file '/home2/db2ins29/pmr/b/datacontd' 40000 )
EXTENTSIZE 32
PREFETCHSIZE 172
FILE SYSTEM
CACHING
;
create large tablespace indexdms
pagesize 8192

Chapter 4. Monitoring 199

managed by database
using ( file '/pmr/amar/indexconta' 20000 , file
'/pmr/amar/indexcontb' 20000 ,
file '/home2/db2ins29/pmr/b/indexcontc' 20000 , file
'/home2/db2ins29/pmr/b/indexcontd' 20000 )
EXTENTSIZE 32
PREFETCHSIZE 128
FILE SYSTEM
CACHING
;

Example 4-37 Table spaces are defined on the following physical/logical volumes

hdisk0:
LV LPs PPs DISTRIBUTION MOUNT POINT
NAME
pmrlv 160 160 00..112..48..00..00 /pmr
paging00 40 40 00..00..40..00..00 N/A
db2lv 20 20 20..00..00..00..00 /db2
data2vgjfslog 1 1 00..00..00..00..01 N/A

hdisk2:
LV LPs PPs DISTRIBUTION MOUNT POINT
NAME
data1vg_jfslogl 1 1 01..00..00..00..00 N/A
homelv 420 420 10..112..111..111..76 /home2
data1vg_raw_lv5 20 20 20..00..00..00..00 N/A
data1vg_raw_lv2 20 20 20..00..00..00..00 N/A
data1vg_raw_lv1 20 20 20..00..00..00..00 N/A
data1vg_raw_lv4 20 20 20..00..00..00..00 N/A
data1vg_raw_lv3 20 20 20..00..00..00..00 N/A

During the creation of the index, table data pages on which the index is to be
created are fetched into the buffer pool if needed. After this, the columns (keys)
on which the index is to be defined are extracted and processed within the
sortheap, if possible. If this is a relatively large index, this process of sorting can
spill into a temporary table space resulting in I/O. Let us create an index and
observe this happening,
200 Best Practices for DB2 on AIX 6.1 for POWER Systems
Slow index create
The relevant dbm and db configuration parameters that were set for this test are
as shown in Example 4-38.

Example 4-38 dbm and db configuration parameters

Enable intra-partition parallelism (INTRA_PARALLEL) = NO

Sort heap thres for shared sorts (4KB) (SHEAPTHRES_SHR) = 5000 Sort
list heap (4KB) (SORTHEAP) = 256
Number of I/O servers (NUM_IOSERVERS) =

1 Number of asynchronous page cleaners (NUM_IOCLEANERS) =

1

The index is created. It takes approximately 16 minutes, as shown in

Example 4-39.

Example 4-39 Index created

time db2 "create index inxall on mystaff ( ID, NAME, DEPT, JOB, YEARS, SALARY,
COMM )"
DB20000I The SQL command completed
successfully. real 16m15.86s
user 0m0.02s
sys 0m0.01s

Let us observe what happened with regards to the index creation. This scenario
is based on DB2 version 9.5, in which the parallelism for the CREATE INDEX
statement is enabled regardless of the setting of the intra_parallel configuration
parameter. From the db2pd -edus output we can see this happening even though
the dbm cfg parameters INTRA_PARALLEL is not enabled. This is because there
are DB2 sub-agents (db2agntdp) performing work on behalf of the coordinator
agent with EDU ID = 1801. See Example 4-40.

Example 4-40 db2pd -edus output

EDU ID TID Kernel TID EDU

Name USR SYS
527 6527 4001859 db2agntdp (SAMPLE ) 0
0.015435 0.006243
6270 6270 5152897 db2agntdp (SAMPLE )0
29.983732 1.296822
6013 6013 5128343 db2agntdp (SAMPLE )0
35.75406 1.534027
3
5756 5756 2662549 db2agntdp (SAMPLE )0
36.241472 1.785477
5499 5499 4968605 db2agntdp (SAMPLE )0
42.88386 2.024181
2

Chapter 4. Monitoring 201

5242 5242 2338831 db2agntdp (SAMPLE )0
34.613903 1.418642
4985 4985 2830441 db2agntdp (SAMPLE )0
35.26929 1.597505
9
4728 4728 4718843 db2agent (idle) 0
0.132892 0.180506
4471 4471 3825687 db2evmgi (DB2DETAILDEADLOCK) 0
0.015022 0.033653
4214 4214 2891985 db2wlmd (SAMPLE) 0
0.014463 0.032677
3957 3957 4952109 db2taskd (SAMPLE) 0
0.001486 0.001270
3700 3700 5029949 db2stmm (SAMPLE) 0
0.003971 0.003357
3443 3443 4407437 db2pfchr (SAMPLE) 0
1.159142 4.693933
3186 3186 2797797 db2pclnr (SAMPLE) 0
0.229199 0.199029
2929 2929 3244285 db2dlock (SAMPLE) 0
0.002543 0.001243
2672 2672 2568379 db2lfr (SAMPLE) 0
0.000031 0.000005
2415 2415 3952751 db2loggw (SAMPLE) 0
0.030844 0.053785
2158 2158 4190425 db2loggr (SAMPLE) 0
0.185102 0.159664
1801 1801 3473637 db2agent (SAMPLE) 0
5.549438 0.737307
1543 1543 4280447 db2resync 0
0.000210 0.000091
1286 1286 4165793 db2ipccm 0
0.019325 0.012484
1029 1029 3788911 db2licc 0
0.000110 0.000158
772 772 1343733 db2thcln 0
0.000733 0.000326
515 515 917711 db2aiothr 0
0.034643 0.061765
2 2 5075121 db2alarm 0
0.005181 0.006202
258 258 667855 db2sysc 0
0.106086 0.067665
202 Best Practices for DB2 on AIX 6.1 for POWER Systems
Let us observe the I/O on the disks. We know that hdisk2 contains /home2, which
also contains the temporary table space. Initially, we see that hdisk2 is quite busy
with regards to the write, as there is not enough room in the sortheap / shared
sort area for the index key sorting. Thereafter, when the sorting operation
finishes, considerable writes are displayed against hdisk0 and hdisk2, which is a
result of writing out the index pages from the temporary table space onto the
disk. Example 4-41 is a formatted output of running iostat while the index
creation was in progress.

After the sort has spilled into the temporary table space in hdisk2, we can see
pages being written to the temporary table space.

Example 4-41 Disk use

Disks: % tm_act Kbps tps Kb_read Kb_wrtn

hdisk0 0.0 0.0 0.0 0 0
hdisk2 0.0 0.0 0.0 0 0
hdisk0 1.5 23.2 2.9 0 48
hdisk2 65.4 1166.1 100.2 440 1968
hdisk0 1.0 5.9 1.5 0 12
hdisk2 98.8 6848.7 276.7 0 13860
hdisk0 0.0 0.0 0.0 0 0
hdisk2 100.0 5076.0 277.0 0 10152
hdisk0 0.0 0.0 0.0 0 0
hdisk2 100.0 6840.0 259.5 0 13680
hdisk0 0.0 0.0 0.0 0 0
hdisk2 100.0 4354.0 283.5 0 8708

This continues for approximately 10 minutes into the index creation, after which
we can see that the sorting has finished and the index itself is being written out to
the index table space on disk, as shown in Example 4-42.

Example 4-42 Index itself is being written out to the index table space
Disks: % tm_act Kbps tps Kb_read Kb_wrtn
hdisk0 0.0 0.0 0.0 0 0
hdisk2 99.9 3264.0 363.3 0 6532
hdisk0 0.0 0.0 0.0 0 0
hdisk2 98.8 3349.0 392.8 0 6472
hdisk0 32.6 210.4 37.1 0 420
hdisk2 100.2 1811.4 184.3 0 3616
hdisk0 60.5 400.0 68.5 0 800
hdisk2 59.0 324.0 55.5 0 648
hdisk0 64.5 400.0 68.5 0 800
hdisk2 73.0 440.0 75.0 0 880
Chapter 4. Monitoring 203
 At the same time, using db2pd with the table space output, we can see this
happening until the entire index is created. It consists of 23584 pages as shown
in Example 4-43.

Example 4-43 db2pd -table spaces output during index creation

Tablespace Configuration:
Address Id Type Content PageSz ExtentSz Auto Prefetch BufID BufIDDisk
FSC NumCntrs MaxStripe LastConsecPg Name
0x070000003040CDE0 0 DMS Regular 8192 4 Yes 4 1 1
Off 1 0 3 SYSCATSPACE
0x070000003040D660 1 SMS SysTmp 8192 32 Yes 32 1 1
On 1 0 31 TEMPSPACE1
0x0700000031F8A120 2 DMS Large 8192 32 Yes 32 1 1
Off 1 0 31 USERSPACE1
0x0700000031F8A940 3 DMS Large 8192 32 Yes 32 1 1
Off 1 0 31
IBMDB2SAMPLEREL N 172 1 1
0x0700000031F8B160 4 DMS Large 8192 32 o
On 4 0 31 DATADMS
0x0700000031F8BDA0 5 DMS Large 8192 32 N 128 1 1
o
On 4 0 31 INDEXDMS

Tablespace Statistics:
Address Id TotalPgs UsablePgs UsedPgs PndFreePgs FreePgs HWM
State NQuiescers
MinRecTim 79872 160 0 79712 160
80000
e 0x0700000031F8BDA0
5
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 160 0 79712 160
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 192 0 79680 192
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 1120 0 78752 1120
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 1824 0 78048 1824
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 2336 0 77536 2336
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 22240 0 57632 22240
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 22848 0 57024 22848
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 23456 0 56416 23456
0x00000000 0 0
0x0700000031F8BDA0 5 80000 79872 23584 0 56288 23584
0x00000000 0 0

204 Best Practices for DB2 on AIX 6.1 for POWER Systems
Correlating to these events we can also see the shared sort usage using
db2mtrk. The same can be noticed using db2pd with the mempools output. Notice
that in Example 4-44 the shared sort increases to 6.8 M. This is roughly the
number of sub-agents multiplied by the sortheap allocated by each of them plus
a little bit extra for the overflow.

Example 4-44 Shared sort usage using either db2mtrk

bph (S16K) bph (S8K) bph (S4K) shsorth lockh dbh

576.0K 448.0K 384.0K 0 640.0K 34.1M
576.0K 448.0K 384.0K 0 640.0K 34.1M
576.0K 448.0K 384.0K 0 640.0K 34.1M
576.0K 448.0K 384.0K 6.3M 640.0K 34.1M
576.0K 448.0K 384.0K 6.8M 640.0K 34.1M
576.0K 448.0K 384.0K 6.8M 640.0K 34.1M
576.0K 448.0K 384.0K 6.8M 640.0K 34.1M

Faster index create

For this case we have made the following changes to the scenario. The script
used to make the changes is shown in Example 4-45.

Example 4-45 Script

db2 update db cfg for sample using SHEAPTHRES_SHR 300000

db2 update db cfg for sample using SORTHEAP 40000
db2 update db cfg for sample using NUM_IOSERVERS AUTOMATIC
db2 update db cfg for sample using NUM_IOCLEANERS
AUTOMATIC db2 alter tablespace datadms PREFETCHSIZE
automatic
db2 alter tablespace datadms NO FILE SYSTEM CACHING

Let us see what happens when attempting to create the same index. The index is
created and it takes approximately four minutes. This is an improvement by a
factor of four over the previous index creation. See Example 4-46.

Example 4-46 Create the same index

[599] db2ins29@eva88 /home2/db2ins29/pmr/b> time db2 "create index

inxall on mystaff ( ID, NAME, DEPT, JOB, YEARS, SALARY, COMM )"
DB20000I The SQL command completed successfully.

real 4m22.45s
user 0m0.01s
sys 0m0.01s

Chapter 4. Monitoring 205

As per the previous case, let us observe what happened with regards to the fast
index creation. The db2pd -edus output looks similar to the output. The difference
is that there are more prefetchers. This can result in better reads, especially
when the data is spread across one or more file systems, as data can be read in
parallel. Again, we see the sub-agents doing the majority of the work in terms of
USR CPU usage. Likewise the converse applies for writing the data using page
cleaners. See Example 4-47.

Example 4-47 db2pd -edus output

EDU ID TID Kernel TID EDU Name USR SYS

8740 8740 4718603 db2agntdp (SAMPLE ) 0 0.002413 0.000842
8483 8483 3932375 db2agnta (SAMPLE) 0 31.873649 0.912463
8226 8226 4223029 db2agnta (SAMPLE) 0 31.151058 0.882912
7969 7969 2977915 db2agnta (SAMPLE) 0 33.142749 0.924740
7712 7712 3465319 db2agnta (SAMPLE) 0 29.103719 0.903962
7455 7455 4280451 db2agnta (SAMPLE) 0 32.160521 0.891415
7198 7198 3473641 db2agnta (SAMPLE) 0 29.682801 0.876035
6941 6941 4165797 db2agent (idle) 0 0.040437 0.057003
6684 6684 5128345 db2evmgi (DB2DETAILDEADLOCK) 0 0.005699 0.011232
6427 6427 2662551 db2wlmd (SAMPLE) 0 0.005559 0.011234
6170 6170 4968607 db2taskd (SAMPLE) 0 0.001373 0.001133
5913 5913 2338835 db2stmm (SAMPLE) 0 0.002828 0.002174
5656 5656 1990779 db2pfchr (SAMPLE) 0 0.057185 0.404891
5399 5399 1986759 db2pfchr (SAMPLE) 0 0.082485 0.435514
5142 5142 4034585 db2pfchr (SAMPLE) 0 0.315293 0.909635
4885 4885 3825693 db2pfchr (SAMPLE) 0 0.408819 1.102220
4628 4628 4190429 db2pclnr (SAMPLE) 0 0.027033 0.022178
4371 4371 4419759 db2pclnr (SAMPLE) 0 0.027504 0.022568
4114 4114 4952113 db2pclnr (SAMPLE) 0 0.025872 0.021212
3857 3857 3788917 db2pclnr (SAMPLE) 0 0.025435 0.020776
3600 3600 1343739 db2pclnr (SAMPLE) 0 0.025373 0.020968
3343 3343 5075127 db2pclnr (SAMPLE) 0 0.025684 0.021113
3086 3086 917717 db2pclnr (SAMPLE) 0 0.026580 0.021514
2829 2829 5161055 db2dlock (SAMPLE) 0 0.000968 0.000484
2572 2572 3952759 db2lfr (SAMPLE) 0 0.000074 0.000026
2315 2315 3629081 db2loggw (SAMPLE) 0 0.011805 0.026484
2058 2058 4554979 db2loggr (SAMPLE) 0 0.057497 0.053045
1801 1801 5173331 db2agent (SAMPLE) 0 5.522430 1.000394
1543 1543 667889 db2resync 0 0.000154 0.000044
1286 1286 5029953 db2ipccm 0 0.005281 0.003560
1029 1029 4407441 db2licc 0 0.000115 0.000173
772 772 2797801 db2thcln 0 0.000073 0.000028
515 515 3244033 db2aiothr 0 0.022980 0.044339
2 2 2568383 db2alarm 0 0.001259 0.001606
258 258 2289899 db2sysc 0 0.093105 0.022703

Let us observe the I/O on the disks. In the previous case there is considerable
write activity against hdisk2, which is spilling into the temporary table space. In
this case however, there is considerable read activity against both hdisk0 and
hdisk2, which is where the table data reside. See Example 4-48 on page 207.
206 Best Practices for DB2 on AIX 6.1 for POWER Systems
Example 4-48 Observing the I/O on the disks
Disks: % tm_act Kbps tps Kb_read Kb_wrtn
hdisk0 0.0 0.0 0.0 0 0
hdisk2 0.0 0.0 0.0 0 0
hdisk0 38.9 11732.0 310.3 23552 0
hdisk2 62.3 11477.0 302.4 23040 0
hdisk0 46.0 11520.0 302.5 23040 0
hdisk2 51.5 11584.0 294.5 23168 0
hdisk0 43.0 11123.0 297.2 22232 0
hdisk2 51.5 11090.9 289.7 22168 0

Unlike the previous case in which this continued for 10 minutes, in this case this
activity only last approximately 3 minutes after which sorting has finished and the
index itself is being written out to the index table space on disk. Notice how the
disk use is significantly higher and likewise with the throughput of the number of
bytes being written. This can be attributed to the greater number of prefetchers.
In the previous example this action took 6 minutes as even though the sorting
had been performed, the pages were required to be read from the temporary
table space into DB2's memory and then written out to disk. In this example, the
operation was much quicker because of the greater number of page cleaners
and the fact that the pages to be written out are already in memory. See
Example 4-49.

Example 4-49 Observing the I/O on the disks

Disks: % tm_act Kbps tps Kb_rea Kb_wrtn
d
hdisk0 89.0 1208.0 151.0 0 2416
hdisk2 94.5 1452.0 181.5 0 2904
hdisk0 90.0 1204.0 150.5 0 2408
hdisk2 84.5 1412.0 163.0 0 2824
hdisk0 88.1 1248.8 156.1 0 2496
hdisk2 99.1 1450.9 184.1 0 2900
hdisk0 99.4 1336.5 167.1 0 2688
hdisk2 91.0 1330.5 170.5 0 2676
Chapter 4. Monitoring 207
 At the same time, using db2pd with the table space output, we can see this
happening until the entire index is created. It consists of 23584 pages. See
Example 4-50.

Example 4-50 db2pd -table spaces output during the index creation

Tablespace Configuration:
Address Id Type Content PageSz ExtentSz Auto Prefetch BufID BufIDDisk
FSC NumCntrs MaxStripe LastConsecPg Name
0x0700000032A9F700 0 DMS Regular 8192 4 Yes 4 1 1
Off 1 0 3 SYSCATSPACE
0x0700000031F8A0A0 1 SMS SysTmp 8192 32 Yes 32 1 1
On 1 0 31 TEMPSPACE1
0x0700000031F8E920 2 DMS Large 8192 32 Yes 32 1 1
Off 1 0 31 USERSPACE1
0x0700000031F8F140 3 DMS Large 8192 32 Yes 32 1 1
Off 1 0 31
IBMDB2SAMPLEREL
0x0700000031F78100 4 DMS Large 8192 32 Yes 128 1 1
Off 4 0 31 DATADMS
0x0700000031F787A0 5 DMS Large 8192 32 N 128 1 1
o
On 4 0 31 INDEXDMS

Tablespace Statistics:
Address Id TotalPgs UsablePgs UsedPgs PndFreePgs FreePgs HWM
State MinRecTim NQuiescers
e
0x0700000031F787A0 5 80000 79872 160 0 79712 160
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 160 0 79712 160
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 448 0 79424 448
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 3520 0 76352 3520
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 6112 0 73760 6112
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 17792 0 62080 17792
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 20800 0 59072 20800
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 23584 0 56288 23584
0x00000000 0 0
0x0700000031F787A0 5 80000 79872 23584 0 56288 23584
0x00000000 0 0

208 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Correlating to the events we can also see the shared sort usage using either
db2mtrk or db2pd with the mempools option. The same can be noticed using
db2pd with the mempools output. Notice the shared sort increases to
approximately 900 MB. This is a little bit less than the number if sub-agents
multiplied by the sortheap allocated for each one, that is, 6x 40000 x 4096 =
983040000.

Example 4-51 db2pd with the mempools output

bph (S32K) bph (S16K) bph (S8K) bph (S4K) shsorth lockh
832.0K 576.0K 448.0K 384.0K 0 640.0K
832.0K 576.0K 448.0K 384.0K 141.4M 640.0K
832.0K 576.0K 448.0K 384.0K 361.7M 640.0K
832.0K 576.0K 448.0K 384.0K 910.9M 640.0K
832.0K 576.0K 448.0K 384.0K 910.9M 640.0K
832.0K 576.0K 448.0K 384.0K 508.1M 640.0K
832.0K 576.0K 448.0K 384.0K 508.1M 640.0K
832.0K 576.0K 448.0K 384.0K 508.1M 640.0K
832.0K 576.0K 448.0K 384.0K 508.1M 640.0K
832.0K 576.0K 448.0K 384.0K 508.1M 640.0K
832.0K 576.0K 448.0K 384.0K 910.9M 640.0K
832.0K 576.0K 448.0K 384.0K 508.1M 640.0K
832.0K 576.0K 448.0K 384.0K 2.1M 640.0K
832.0K 576.0K 448.0K 384.0K 2.1M 640.0K
832.0K 576.0K 448.0K 384.0K 2.1M 640.0K

We can see that changing a few parameters at the database configuration and
the table space level can lead to a significant improvement in the amount of time
it takes to create an index. By optimizing the database to take full advantage of
the multi-threaded architecture and exploiting the memory available, it can
alleviate the various bottlenecks that otherwise are left unchanged.

4.8.3 Case 3: Alleviating I/O bottleneck

An OLTP application was tested on the test machine and deployed on the
production server. Production server is an 8-way Power 6 processor with 128 GB
RAM, while the test server runs on a 4-way Power3 processor with 8 GB RAM.
The production server is attached to DS3400 Storage Server with 96 spindles.

Despite production servers having a relatively high configuration compared to the

test server, the throughput was almost the same on both production and test
server.
Chapter 4. Monitoring 209
 Analysis
nmon and DB2 Snapshot were collected during the workload run. Looking at the
CPU use seen in Example 4-52 and Figure 4-25, it seems we are hitting an I/O
bottleneck. The average I/O wait (measured by wa) is 50%. Despite 100
connections, the run queue is low, at 3 threads.

Example 4-52 CPU use

kthr memory page faults cpu
====================================================================
===========
r b p avm fre fi fo pi po fr sr in sy cs us sy id wa
3 0 0 4966174 21500 129 0 4 0 0 0 0 4480 106056 17 7 27 48

Figure 4-25 CPU total

210 Best Practices for DB2 on AIX 6.1 for POWER Systems
Let us take a look at statistics related to disk activity in nmon. There are eight
hdisks that are fairly active: hdisk23, hdisk24, hdisk25, hdisk26, hdisk27,
hdisk28, hdisk29 and hdisk30. See Figure 4-26.

Figure 4-26 Disk%busy

Chapter 4. Monitoring 211
All disks, except for hdisk24, are characterized by read activity, as shown in
Figure 4-27.

Figure 4-27 Disk Read KB/s

hdisk24 has relatively high write activity, as seen in Figure 4-28.

Figure 4-28 Disk write KB/s

212 Best Practices for DB2 on AIX 6.1 for POWER Systems
The I/O activity looks balanced across the adapters, as shown in Figure 4-29.

Figure 4-29 The I/O activity looks balanced across the adapters

So far we have not been able to determine any obvious issue with I/O. Let us see
what the DB2 Key Performance Indicators suggest. Is it that DB2 is driving many
physical I/Os? Let us map the LUNs to file systems. See Example 4-53.

Example 4-53 LUNs to file systems

/usr/sbin/lspv -l hdisk23
hdisk23:
LV NAME LPs PPs DISTRIBUTION MOUNT
POINT
DATALV7 900 900 00..819..81..00..00 /DATA7

hdisk24 maps to /LGSHR

hdisk25 maps to /DATA1
hdisk26 maps to /DATA2
hdisk27 maps to /DATA3
hdisk28 maps to /DATA4

hdisk29 maps to /DATA5

hdisk30 maps to /DATA6
Chapter 4. Monitoring 213
Next, let us map the file systems to DB2 table space containers and logs, as
shown in Example 4-54.

Example 4-54 DB2 table space containers and log

SELECT TABLESPACE_ID,
TABLESPACE_NAME,CONTAINER_NAME FROM
TABLE(SNAPSHOT_CONTAINER('PRDDB',-1)) AS T

TABLESPACE_ID TABLESPACE_NAME CONTAINER_NAME

------------------------------------------- --------------------------
6 TS_TXM /DATA1/PRDDB1
6 TS_TXM /DATA2/PRDDB1
6 TS_TXM /DATA3/PRDDB1
6 TS_TXM /DATA4/PRDDB1
6 TS_TXM /DATA5/PRDDB1
6 TS_TXM /DATA6/PRDDB1
6 TS_TXM /DATA7/PRDDB1

db2 get db cfg for prddb1 | grep –i ““Path to log files” Path to

log files = /LGSHR

TS_TXM table space is using the DATA[1-7] file systems as containers, while
/LGSHR is where the transaction logs of the database are stored. This explains
why we saw relatively higher percentage of writes on hdisk24.
Let us see what tables are active in TS_TXM table space. See Example 4-55.
The number of rows read is 0. There are no tablescans occurring against the
tables.

Example 4-55 Tables that are active in TS_TXM table space

SELECT TABLE_NAME, ROWS_READ FROM TABLE(SNAPSHOT_TABLE('TPCC',-1))

WHERE TABLE_NAME IN (SELECT TABNAME FROM SYSCAT.TABLES WHERE
TBSPACEID=6)

TABLE_NAME ROWS_READ
------------------
-----------------------------------------------------------------------
--
CUSTOMER 0
ORDERS 0
WAREHOUSE 0
DISTRICT 0
NEW_ORDER 0
STOCK 0
 HISTORY 0
ORDER_LINE 0

214 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Example 4-56 shows that the bufferpool hit ratio is quite good.

Example 4-56 Buffer pool Hit Ratio

SELECT BP_NAME, TOTAL_HIT_RATIO_PERCENT, DATA_HIT_RATIO_PERCENT,
INDEX_HIT_RATIO_PERCENT
FROM SYSIBMADM.BP_HITRATIO

BP_NAME
TOTAL_HIT_RATIO_PERCENT DATA_HIT_RATIO_PERCENT INDEX_HIT_RATIO_PERCENT
-------------------------------------------------------------------------------
------------------------------------------------- -----------------------
---------------------- ----------------------- IBMDEFAULTBP
99.39 97.90 99.90

Let us check the Read Response time in the Snapshot. See Example 4-57.
Synchronous read time is important because it blocks the agents and prevents
them from doing useful work. This shows up as more idle time in vmstat.
Example 4-57 Average read time of 72 ms points to poor I/O response time

SELECT BP_NAME, AVERAGE_SYNC_READ_TIME_MS FROM SYSIBMADM.BP_READ_IO

BP_NAME AVERAGE_SYNC_READ_TIME_MS
----------------------------------------------------------------------------- IBMDEFAULTBP

72

To get details about I/O response time, let us monitor iostat –D at regular
intervals. We are constantly hitting sqfull condition for all LUNs except hdisk24.
sqfull indicates the number of times the service queue becomes full per second.
That is, the disk is not accepting any more service requests. On an average,
every I/O transfer is spending 14 milliseconds in the wait queue. There might be
two possible reasons: Either we have exhausted on the bandwidth or
queue_depth of the device is not set properly. Example 4-58 shows hdisk30 use.

Example 4-58 hdisk30 use

hdisk30 xfer: %tm_act bps tps bread bwrtn
90.1 3.5M 854.1 3.5M 0.0
read: rps avgserv minserv maxserv timeouts fails
854.1 3.0 0.1 28.3 0 0
write: wps avgserv minserv maxserv timeouts fails
0.0 0.0 0.0 0.0 0 0
queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull
14.0 0.0 234.6 0.8 0.2 806.4
Chapter 4. Monitoring 215
 Next, let us check the queue depth of the devices:
lsattr -E -l hdisk30 -a queue_depth
queue_depth 3 Queue DEPTH True

They have been left at default of 3. The Storage Administrator informs us that
there are 12 physical disks (spindles) under each LUN. The rule of thumb is that
at least three outstanding I/Os can wait on each physical disk device to keep it
near 100% busy. Because seek optimization is employed in current storage
subsystems, having more than three outstanding I/O can improve
throughput/response time for I/Os. The queue_depth of the LUNs was set to five
times the number of spindles. The num_cmd_elems was also adjusted for the
adapters, so that it is at least equal to the sum of queue_depth of all LUNs
attached to it.
chdev –l hdisk23 -a queue_depth=60
chdev–lfcs0–anum_cmd_elems=120

sqfull was monitored and there were no sqfull observed for the devices. The
Average Buffer pool Synchronous Read time reduced to 11.5 milliseconds.

This helped improve the throughput of the OLTP workload by 3x.

4.8.4 Case 4: Memory consumed by the AIX file system cache

The production server hosts Sales Data Warehouse was a well performing
system until recently. Performance of the warehouse has degraded with long
turnaround time for any workload, impacting business. At times, the server has
become unresponsive and had to be rebooted.

Analysis
High level of paging activity is noticed in vmstat when the response time drops.
Take note of the pi and po columns. See Example 4-59.

Example 4-59 vmstat output

System configuration: lcpu=4 mem=16384MB ent=2.00
kthr memory page faults cpu
----- ----------- ------------------------ ------------ -----------------------
r b avm fre re pi po fr sr cy in sy cs us sy id wa pc ec
0 28 3010156 1552 0 38 1140 1152 1152 0 543 604 11393 7 16 14 62 0.25 24.6
8 14 3011227 1552 0 70 605 512 512 0 469 444 5595 3 18 29 50 0.22 22.3
1 10 3012633 1552 0 35 740 768 768 0 476 512 4564 4 11 38 47 0.16 16.0
0 33 3013989 1552 0 50 724 640 640 0 433 316 11996 5 19 28 48 0.25 24.8
0 15 3015311 1552 0 46 707 704 704 0 508 559 6221 4 13 36 47 0.19 18.6
0 14 3016184 1552 0 26 463 448 448 0 394 292 5644 3 10 46 40 0.15 14.8
0 10 3017089 1552 0 24 475 448 448 0 431 360 4900 3 9 56 33 0.12 12.4
216 Best Practices for DB2 on AIX 6.1 for POWER Systems
Let us use lsps to determine the paging space use details. We do have adequate
paging space around. However, the “%Used” in lsps is increasing over time. See
Example 4-60.

Example 4-60 lsps -a

Page Space Physical Volume Volume Group Size %Used Active Auto Type
paging01 hdisk9 work2vg 32768MB 20 yes yes lv
paging03 hdisk8 home2vg 24576MB 20 yes yes lv
paging02 hdisk7 home1vg 24576MB 20 yes yes lv
paging00 hdisk6 systemvg 16896MB 20 yes yes lv
hd6 hdisk0 rootvg 4096MB 20 yes yes lv

It is important to understand the effect of AIX tunables, minperm and maxperm,

with respect to paging. As mentioned in Chapter 2, “AIX configuration” on
page 35, protecting computational memory is important for DB2. In other words,
we need to favor computational pages over file system cache pages (client and
persistent pages).
› Minperm
If percent of real memory occupied by file pages falls below this level, VMM
steals both file and computational pages. Hence, a low value (3%) is
recommended for DB2.
› Maxperm
If percent of real memory occupied by file pages increases beyond this level,
VMM steals only file pages. The recommended value for this is 90%.

Let us get more insights the memory usage. The -v option of vmstat displays the
percentage of real memory being used by other categories of pages

vmstat -v was collected and monitored at regular intervals. The minperm and
maxperm are set, as per recommendations. However, numperm and numclient is
high at 70%. This suggests that the file system cache is occupying 70% of the
real memory. See Example 4-61.

Example 4-61 vmstat -v

/usr/bin/vmstat -v 16318448
memory pages
15690369 lruable pages
144213 free pages
2 memory
pools 1060435 pinned
pages
80.0 maxpin percentage
3.0 minperm percentage <<- system’s minperm setting
90.0 maxperm percentage <<- system’s maxperm setting
Chapter 4. Monitoring 217
 70.9 numperm percentage <<- % of memory
having non-computational pages
11138217 file pages <<- No of non-computational pages
0.0 compressed percentage
0 compressed pages
70.5 numclient percentage <<- % of memory containing non-comp
client pages
90.0 maxclient percentage <<- system ‘s maxclient setting
11074290 client pages <<- No of client pages
0 remote pageouts scheduled
36 pending disk I/Os blocked with no pbuf 3230
paging space I/Os blocked with no psbuf 21557
filesystem I/Os blocked with no fsbuf
1 client filesystem I/Os blocked with no fsbuf
0 external pager filesystem I/Os blocked with no fsbuf
0 Virtualized Partition Memory Page Faults
0.00 Time resolving virtualized partition memory page faults

nmon also displays the memory statistics. The MEMNEW tab in the nmon
spreadsheet displays the details about usage of memory. See Figure 4-30.

Figure 4-30 nmon output

218 Best Practices for DB2 on AIX 6.1 for POWER Systems
Next, let us check which process is consuming the file system cache. We use
svmon to list the top 10 consumers of file system cache.
› -c option of svmon provides details about client memory
› -f about persistent
› -w about computational memory

Example 4-62 svmon output

$ svmon -c -Pt 10
-----------------------------------------------------------------------
--------
Pid Command Inuse Pin Pgsp Virtual 64-bit Mthrd
16MB
160752 db2sysc 8859432 0 0 Y
0
Y N =

PageSize Inuse Pin Pgsp Virtual

s 4 KB 10887 0 0 0
m 64 KB 0 0 0 0

Vsid Esid Type Description PSize Inuse Pin

Pgsp Virtual
585cb2 - clnt /dev/data_lv8:4098 s 1476572 0 -
-
779aec - clnt /dev/data_lv7:33 s 1476400 0 -
-
e815d3 - clnt /dev/data_lv3:33 s 1476882 0 -
-
822107 - clnt /dev/data_lv6:33 s 1476360 0 -
-
21a040 - clnt /dev/data_lv4:33 s 1476770 0 -
-
8edd1f - clnt /dev/data_lv5:33 s 1476172 0 -
-
780ef3 10 clnt text data BSS heap, s 34 0
- -
/dev/homeDSlv:542222
3c907b 9fffffff clnt USLA text,/dev/hd2:94283 s 14 0
- -
6900e - clnt /dev/flat_lv:12343 s 6 0 -
-
1e2c3e - clnt /dev/flat_lv:12345 s 6 0 -
-
d2efa7 - clnt /dev/flat_lv:12684 s 3 0 -
-

Chapter 4. Monitoring 219

 d4f3ab - clnt /dev/hd9var:620 s 3 0
- -
381072 - clnt /dev/hd2:78707 s 2 0
- -
cb0195 - clnt /dev/flat_lv:12461 s 2 0 -
-
db9db4 - clnt /dev/flat_lv:12652 s 1 0 -

db2sysc is at the top of the list. The InUse column displays the total number of
client pages DB2 is using. It alone accounts for 80% of numclients (8859432
pages out of 11074290 client pages).

If you want to check how much space db2sysc is occupying in the file system
cache, you can pass db2sysc’s pid as an argument (svmon -c –P <db2sysc-pid>)

Why is DB2 using so much of file system cache? Either it is the DB2 logs (less
likely, as this is a data warehouse workload and logging activity is minimal) or file
system caching is ON for DB2 Tablespace. File System Caching is not
recommended for DB2 containers, except for cases where your table space has
LOB data. As DB2 already buffers recently used data in buffer pool, having file
system caching ON unnecessarily creates two levels of buffering.

In case you are curious, this is how we can determine files db2sysc is using from
the svmon output. The Description column points to the location and inode of the
file. Let us map /dev/data_lv8:4098 to file. See Example 4-63.

Example 4-63 lsfs output

lsfs /dev/data_lv8
Name Nodename Mount Pt VFS Size Options
Auto Accounting
/dev/data_lv8 -- /data08 jfs2 65536000 rw
no no

# find /data08 -inum 4098

/data08/ts_work/con08.dat
220 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Next, check if file system caching is ON for any DB2 table space using table
space snapshot.
db2 get snapshot for tablespaces on <dbname>

Example 4-64 Tablespace snapshot

Tablespace name = DATA_TS
Tablespace ID =0
Tablespace Type = Database managed space
Tablespace Content Type = All permanent data.
Regular table space.
Tablespace Page size (bytes) = 4096
Tablespace Extent size (pages) =4
Automatic Prefetch size enabled = Yes
Buffer pool ID currently in use =1
Buffer pool ID next startup =1
Using automatic storage = Yes
Auto-resize enabled = Yes
File system caching = Yes

File system caching was ON for one table space. After disabling file system
cache for the table space, there was no paging observed and that data
warehouse performance was back to normal.

In this scenario, we found that DB2 is the culprit. In case you find non-DB2
processes are consuming file system cache, you’d need to determine the cause
and rectify it.

4.8.5 Case 5: Disk performance bottleneck

In this scenario, a performance bottleneck related to DB2 on AIX using RAID
array disk is encountered when running various workload stress tests. This
section focuses on the approach taken to fix this issue and provides a few hints
on tuning the related parameters. As the OLTP workload is increased, there is
throughput and a 70% increase in response time of the transactions compared to
the baseline numbers captured with half the load.
Chapter 4. Monitoring 221
 Using the following nmon graph on CPU shows consistent CPU wait%, which
shows that the processing is on wait for I/O or disk activity. See Figure 4-31.

Figure 4-31 nmon output

One thing to check for using the DB2 monitor or snapshots is the buffer pool size
usage and hit ratio. Remember, a smaller buffer pool size and lower buffer pool
hit ratio can cause high disk I/O usage, which causes the CPU waits to go higher.
In this case, the setting used for buffer pool was auto tuning and the DB2
monitoring shows that the DB2 buffer pool hits ratio is 99% which ruled out this
being a cause. See Figure 4-32.

Figure 4-32 nmon chart

222 Best Practices for DB2 on AIX 6.1 for POWER Systems
From the nmon chart, we can see that an increasing load results in longer disk
I/O time (and waits). DB2 does both sequential and random reads. But random
reads are done in page sizes, so if you do a read ahead on a random read, the
disk actually takes longer to read the page as it transfers more data. Sequential
read ahead is normally a good thing for sequential reads, because even if you
read one page, the disk reads the rest of it, anticipating your next read request,
and although it slows down the first read request, the speed of the subsequent
reads improves significantly.

DB2 is normally smart, and when it does sequential reads, it does not do them in
small units, but instead reads multiple pages at a time (large I/O). In this case,
these large sequential reads see no benefit from read ahead because the
requested I/O size is probably similar to the disk stripe size used by read ahead.
The OLTP workload in this case, was suspected to be causing such an issue of
large waits due the read requests being random I/O as opposed to sequential.

Based on this, the sequential read ahead AIX setting was tuned and
experimentation lead to a maximum throughput and reduced the response time
such that the negative performance was alleviated. The following AIX settings
are under investigation to fix this issue.
› ioo -o minpageahead=Number
› ioo -o maxpageahead=Number

Note: Prior to AIX5.3 version, use, minpgahead and maxpgahead values can
be used with options -r and -R respectively in the vmtune AIX command. For
details on using these options, see AIX 5L Performance Tools Handbook,
SG24-6039.

Using higher values of maxpgahead can be applied in environments where the

sequential performance of striped logical volumes is of paramount importance.
The initial settings for this scenario were minpageahead as 4 and
maxpageahead as 16. However, in this scenario, this value had a negative
negative effect which merely added to the IO waits.

To fix this issue, the minpgahead value was varied from 0–4 with maxpgahead
varied from 8–16. Subsequent tests were carried out for each of these. The
minpgahead being 0 showed maximum improvement in performance for this
issue. Remember, a value of 0 for minpgahead can be useful for situation’s of
random I/O or workload that has predominantly doing random I/O, as it is totally
opposite to the mechanism of sequential read ahead. As a result, the response
time difference came down to 10% from 60–70% due to an increase in workload
and it resulted in better throughput values.
Chapter 4. Monitoring 223
224 Best Practices for DB2 on AIX 6.1 for POWER Systems
 5

Chapter 5. LPAR considerations

As seen in1.5, “Introduction to PowerVM virtualization” on page 21, logical
partition (LPAR) is a fundamental building block for the PowerVM virtualization
environment. For effective DB2 deployment in the virtualized environment, it is
imperative to understand various LPAR features and how to apply them. The
chosen LPAR features directly influence functional and non-functional
requirements of a datacenter.

In this chapter we discuss at a length about best practices for using the
appropriate LPAR type, applying uncapped micro-partitions, multiple shared CPU
pools, and how to manage resources and prioritize resource allocation among
multiple LPARs hosted on a system.

We cover the following topics in this chapter:

› “LPAR planning” on page 226
› “LPAR profile” on page 227
› “Dynamic logical partitioning” on page 229
› “Dedicated LPAR versus micro-partition” on page 231
› “Uncapped micro-partition” on page 234
› “Micro-partition: Virtual processors” on page 235
› “Multiple shared-processor pools” on page 236
› “LPAR monitoring” on page 238
© Copyright IBM Corp. 2010. All rights reserved. 225
5.1 LPAR planning
Logical partition (LPAR) is an isolated computing domain. Each LPAR needs its
own processing capacity, memory, network connection, storage, and operating
system. POWER Hypervisor™ encapsulates the system hardware resources to
provides levels of abstraction to the LPARs configured on a system. The LPARs
running on a system are completely independent from each other. LPARs can
have differing amounts of resources and in fact can run any supported operating
system (such as AIX, Linux on POWER, and i5/OS®).

The system shown in Figure 5-1 is configured with four LPARs. The LPAR with
AIX 6.1 is dedicated processor LPAR. The rest of the LPARs are shared
processor LPARs. Each LPAR is running variety of operating systems with
differing amount of resources.

CoD
CPUs

Dedicated CP U Virtual CPU Virtual CPU Virtual CPU CoD CPU

Dedicated CP U Virtual CPU Virtual CPU CoD CPU

Virtual CPU CoD CPU
Virtual CPU CoD CPU

CPU CPU CPU CPU CPU

CPU
CPU CPU

Dynamic Spares and

Capacity on Demand
IB M

Dedicated
Physical CPUs

Shared Pool of
Physical CPUs

Figure 5-1 LPAR configurations

When setting up the system, carefully consider the LPAR requirements,

especially the number of planned LPARs with respect to system physical
resources, including the number of configured physical processors in a system
226 Best Practices for DB2 on AIX 6.1 for POWER Systems
 and number of I/O slots. Depending on model type, size, and installed options
such as PowerVM, a POWER6 system can run up to 255 LPARs.For example, if
we configure more LPARs than the number of physical processors in a system, a
few of the LPARs are shared processor LPARs. Similarly, a limited number of I/O
adapters require use of Virtual I/O Servers (VIOSs), because each LPAR needs
storage for installing an operating system, such as rootvg for AIX. More
information about VIOS, and how to apply it in DB2 environment is available in
Chapter 6, “Virtual I/O” on page 241.

After the LPAR is created and the AIX operating system is installed, the user
experience is similar to working on a regular host running AIX. However, there
are new concepts and setup options related to the virtualization that we discuss
in the remainder of the chapter.

5.2 LPAR profile

When a system administrator provisions an LPAR, each LPAR has a set of
attributes associated with it. In general, the attributes establish the amount of
resources (such as processor, memory, I/O) an LPAR can consume. LPAR
resources can be altered even at runtime without requiring a restart of AIX 6L or
DB2, using the dynamic logical partition (DLPAR) facility. Any DLPAR operation
must honor the specific resource limit set in the LPAR profile. For example, if an
LPAR profile dictates that a minimum amount of entitled capacity is 0.50, a
DLPAR operation requesting any lower entitled capacity is rejected with an error.
Chapter 5. LPAR considerations 227
 Figure 5-2 shows a LPAR properties information dialog box for processor
configuration.

Figure 5-2 LPAR processor minimum, desired, maximum attributes

228 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Figure 5-3 shows LPAR properties information dialog box for memory
configuration.

Figure 5-3 LPAR memory minimum, desired, and maximum attributes

Processor and memory resources have minimum, desired, and maximum values
in an LPAR profile. The desired value is used when LPAR is started. If the desired
amount cannot be fulfilled when starting an LPAR, the POWER Hypervisor
attempts to start an LPAR using any amount between the minimum and desired
value. If the minimum amount cannot be fulfilled, LPAR fails to start.

5.3 Dynamic logical partitioning

Dynamic logical partitioning (DLPAR) allows altering partition’s resources while
an LPAR is running, without the need to reboot the system. These partition
resources can be added or removed from the LPAR without requiring a restart of
the LPAR or DB2. The following resources can be dynamically altered:
› Physical processors (for dedicated processor LPARs)
› Virtual processors or entitled capacity (for micro-partitions)
› Memory
› Physical I/O adapters
› Virtual I/O adapters
Chapter 5. LPAR considerations 229
It is an effective tool for dynamic workload management. The traditional workload
management strategy is to size and allocate resources for peak workload. In
most cases, however, it results in under-used resources when workload demand
is not at the peak.

Rather than provisioning resources for peak usage, use the current workload
resources requirement to allocate LPAR resource. When the demand changes,
use the DLPAR facility to add or remove the required resources. For example, if
the current processor consumption decreases, use DLPAR to remove processor
resources from an LPAR. Similarly, when AIX 6.1 or DB2 needs more memory (in
event of an AIX 6.1 paging activity or low DB2 buffer pool hit ratio, for example),
more memory can be added dynamically. Similarly, the resources can be moved
from one LPAR to another.

The DLPAR operation is easily accessible using the Hardware Management

Console graphical interface. Figure 5-4 shows the menu options to perform the
DLPAR operation.

Figure 5-4 Dynamic logical partitioning (DLPAR) menu

Note: If you are using the DLPAR facility to add or remove memory, we
recommend the use of DB2’s self tuning memory management (STMM)
feature for most benefits. The STMM, with the INSTANCE_MEMORY
database manager configuration parameter, and the DATABASE_MEMORY
database configuration parameter (both set to AUTOMATIC) ensures that
memory change in an LPAR is registered with DB2 for optimized memory use.
Additionally, it takes the burden from the database administrator to retune DB2
after a memory change in an LPAR.
230 Best Practices for DB2 on AIX 6.1 for POWER Systems
5.4 Dedicated LPAR versus micro-partition
There are two types of LPAR. Most of the time, questions arise regarding the
correct partition type for the DB2 workload and environment. Besides
performance, there are other factors that contribute to deciding the partition type.

A test was conducted with one, two, and four partitions running simultaneously
for each partition type. All tests used direct attached I/O with properly scaled DB2
data. In essence, the test contained a set of six performance runs: two runs for a
given number of partitions, each time changing only the partition type between
dedicated and shared-processor partitions. The partition type change is a
nondestructive change (no data loss). The only requirement is that the partition
be restarted.

Figure 5-5 shows the results of these tests.

Figure 5-5 Result of the tests

Chapter 5. LPAR considerations 231
5.4.1 LPAR throughput
As shown in Figure 5-6, there is no difference between the types of partitions.

Dedicated processor vs. Shared processor LPAR

Normalized cumulative throughpu

1 LPAR 2 LPARs 4 LPARs

(8 processors/LPAR) (4 processors/LPAR) (2 processors/LPAR)

Dedicated processor LPAR Shared processor LPAR

Locally attached I/O Locally attached I/O

Figure 5-6 Compare dedicated and shared-processor partitions

Performance of the shared processor partition (for example the cumulative

throughput) is identical to the dedicated partition, for one, two, and four partitions.
This behavior is validation of the fact that POWER Hypervisor uses an efficient
scheduling algorithm for the shared processor partitions with reasonably large
processor entitlement. For such shared processor partitions, POWER Hypervisor
attempts to schedule on the same physical cores each time, similar to dedicated
processor partition.

5.4.2 LPAR scalability

Another interesting observation from the same data is depicted in Figure 5-7 on
page 233. There is no partitioning scalability overhead for either type of partition.
That is, the amount of work done by an eight processor dedicated processor
partitions (or shared processor partitions with 8.00 entitled capacity with eight
virtual processors) is twice that of the four processor partition, which, in turn, is
twice that of the two processor partition.
232 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Normalized average throughput / LPAR

1 LPAR 2 LPARs 4 LPARs

(8 processors/LPAR) (4 processors/LPAR) (2 processors/LPAR)

Dedicated processor LPAR Shared processor LPAR

Locally attached I/O Locally attached I/O

Figure 5-7 Dedicated processor and shared processor LPAR scalability

It is more cost-effective to reduce unused processing capacity within a partition

by appropriately sizing a partition, which results in improved use at a system
level. A general rule of thumb is not to waste processing capacity within a
partition. Thus, the recommendation is to use a shared processor partition for
DB2 deployments. The shared processor partition improves processor use for
the whole system and offers flexible scaling using the uncapped feature when the
workload demand exceeds the currently assigned processor entitlement.

5.4.3 Discussion
Beginning with POWER6 processor architecture-based servers, there is an
option to donate unused dedicated processor partition processing capacity. This
feature eliminates unused processing cycles in a dedicated processor partition,
further solidifying PowerVM virtualization offering.

Strictly speaking from a performance point of view, theoretically, because of

processor and memory affinity, a dedicated partition must offer optimal
performance. However, it has been found that this is not the case for DB2
Chapter 5. LPAR considerations 233
 workloads. That is, there is no performance difference between dedicated and
shared-processor partitions for the system under test.

With the same performance characteristics for both dedicated- and

shared-processor partitions, consider other aspects and differences between the
partition types. For a dedicated partition, processor assignment is in increments
of one whole processor. Therefore, idle processor capacity often occurs within a
dedicated partition. For example, a dedicated partition with three processors that
runs DB2 and consumes the equivalent to 2.10 processors (that is, vmstat shows
70 percent partition use) results in 0.90 processor wasted capacity unless shared
dedicated capacity for this LPAR is turned on.

The dedicated partition allows adding or removing processor resources at run

time, without restarting AIX or DB2. The only way to alter dedicated partition
processor allocation is by using a dynamic LPAR operation. Dynamic LPAR is
relatively slower (with a latency of a few seconds, depending on the workload)
than the near-instant entitled capacity change that is offered by an uncapped
shared-processor partition. A shared-processor partition offers many
advantages, including suitability for dynamic, unpredictable workloads, effective
resource use, and easy workload management.

Tip: Whenever possible use a shared processor LPAR for running DB2. The
shared processor LPAR (micro-partition) offers the same level of performance
as a dedicated processor LPAR for most kinds of DB2 workloads. Besides, the
shared processor LPAR offers many other workload management features
and improves overall server use.

5.5 Uncapped micro-partition

This is one of the primary distinguishing features between dedicated and
shared-processor partitions. The dedicated-processor partition is capped by
definition. That is, a dedicated-processor partition cannot use more than its
allocated processors, even if idle processing capacity is available on the system.
When applied appropriately, uncapped shared-processor partition is effective in
dealing with highly dynamic or unpredictable workloads. The uncapped weight is
used to arbitrate free-pool capacity allocation among competing uncapped
shared-processor partitions.

As an example, consider a scenario where a multitier infrastructure consists of a

DB2 server as a back-end tier and IBM WebSphere as an application server.
Each is hosted in an individual, uncapped shared-processor partition of a System
p server. Assume that the DB2 partition is a high-priority partition. Therefore, the
partition's uncapped weight is set higher than the WebSphere partition. At peak
234 Best Practices for DB2 on AIX 6.1 for POWER Systems
 demand, both DB2 and WebSphere exceed the entitled capacity and are allowed
to use free-pool processing capacity (if available). However, the DB2 partition
gets more free-pool processing cycles than WebSphere.

5.6 Micro-partition: Virtual processors

The use of micro-partitions requires setting the number of virtual processors to
an appropriate value. Under-configuring the number of virtual processors can
lead to wasted processing capacity or might not allow the use of free-pool
processing cycles (the uncapped partition's maximum entitled capacity is limited
to the number of virtual processors). Over-configuring of virtual processors is not
a problem, because, by default, AIX does not schedule unused virtual
processors. (refer to AIX schedo command configuration parameter vpm_xvcpus)

A test was performed to measure virtual-processor folding effectiveness by

varying the number of virtual processors for one of the shared-processor
partitions from the four-partition scenario listed in Figure 5-5 on page 231. As
expected, a relatively higher number of virtual processors did not have any
negative performance impact. The partition had an entitled-processor capacity of
1.85. Three runs were performed. The first test involved the number of virtual
processors being set to round up the entitled capacity (1.85 to 2) Subsequent
tests configured a much higher number of virtual processors.

Based on the recorded results on the setup, a recommendation for the DB2
environment, for capped partitions, is to set the number of virtual processors
equal to the rounded-up entitled capacity.

For an uncapped partition, we suggest the number of virtual processor to be

twice the round up value of the current entitlement, but not to exceed the total
number of physical processors configured in a server. A 16-way server, hosting
an uncapped shared processor LPAR with an entitled capacity of 1.85 must have
four virtual processors, whereas another uncapped shared processor LPAR (on
the same system) with an entitled capacity of 8.50 must have 16 virtual
processors.

This configuration for uncapped partitions ensures that there are sufficiently
configured virtual processors to allow the use of the free-pool processing
capacity. Carefully tune uncapped weight to assist fair distribution of the free pool
capacity.
Chapter 5. LPAR considerations 235
 Tip: For DB2 environment use the following general guideline to configure
virtual processors for shared processor LPAR.

For capped shared processor LPAR, set the number of virtual processor to be
equal to the rounded-up number of the entitled capacity.

For uncapped shared processor LPAR, set the number of virtual processor to
be equal to the round-up of minimum (2 X entitled capacity, number of pyhsical
processors in a system)

5.7 Multiple shared-processor pools

POWER6 processor-based systems support Multiple Shared-Processor Pools
(MSPPs). Unlike the previous mechanism, where all micro-partitions belong to
system wide default pool (called a physical shared processor pool), this new
capability allows a systems administrator to create a set of micro-partitions in a
separate, user-defined pool.

Note: For more information about multiple shared-processor pools see the
IBM Redbooks publication PowerVM Virtualization on IBM System p:
Introduction and Configuration Fourth Edition, SG24-7940.

MSPPs use processor capacity from the physical shared processor pool. There
can only be one physical shared processor pool in the system. Each
shared-processor pool has an associated entitled pool capacity, which is
consumed by the set of micro-partitions belonging in that shared-processor pool.

The micro-partitions within a shared-processor pool are guaranteed to receive

their entitled capacity. In addition, unused processor cycles within their
shared-processor pool are harvested and are redistributed to eligible
micro-partitions within the same shared-processor pool.

The source of additional processor cycles can be the reserved pool capacity,
which is processor capacity that is specifically reserved for this particular
shared-processor pool, but not assigned to any of the micro-partitions in the
shared-processor pool.

When the set of micro-partitions in a shared-processor pool are heavily loaded,

they can consume additional processor capacity (assuming they are uncapped)
from outside their shared-processor pool up to a defined maximum (the
maximum pool capacity). Processor capacity distributed in this way has been
ceded by underused sets of micro-partitions in their shared-processor pool.
236 Best Practices for DB2 on AIX 6.1 for POWER Systems
There is a default shared-processor pool (shared-processor pool0) that is used
for all micro-partitions if the system administrator has not created other
shared-processor pools. The shared-processor pool0 has properties that ensure
operational consistency with POWER6 processor-based micro-partitioning and
the physical shared processor pool.

Figure 5-8 shows an architectural overview of a POWER6-based server with two

shared-processor pools defined.

Dedicated-processor
partitions Set of micro-partitions Set of micro-partitions

AIX V6.1

AIX V5.3

AIX V6.1

AIX V6.1

AIX V5.3
AIX V5.3
AIX V5.2

AIX V5.3

AIX V6.1

Linux

Linux
0.75 CPU

0.75 CPU

0.85 CPU
0.5 CPU

1.6 CPU

0.5 CPU

0.6 CPU

0.5 CPU
2 CPU 3 CPU 1 CPU

Shared-Processor Pool 0 Shared-Processor Pool 1

POWER Hypervisor

Dedicated processors Physical shared-processor pool

Figure 5-8 POWER6-based server with MSPPs defined

There can be up to 64 MSPPs in total. With the default MSPP, there can be an
additional 63 MSPPs that are user-defined.

Workload management using the uncapped feature of micro-partition becomes

challenging if a system has an excessive number of uncapped micro-partitions.
The uncapped weight offers limited tuning options for distributing free pool
capacity. Additionally, managing uncapped weight across dozens of
micro-partitions is challenging, too. To address this, a shared processor pool can
be defined with a relatively small number of micro-partitions belonging to it. This
method offers a more granular processor resource management.

If a system is hosting a variety of applications and workloads, we suggest a

separate shared-processor pool for application stack containing DB2 workloads.
For example, in the case of a multi-tier application stack, all related stack
Chapter 5. LPAR considerations 237
 components are part of the same shared processor pool for granular processor
resource management. Also, if the VIOS is an uncapped micro-partition, and is
predominantly servicing DB2 I/O requests, the VIOS is recommended to be part
of the same shared processor pool as well.

5.8 LPAR monitoring

The standard AIX monitoring tools and interfaces are enhanced for the
virtualization environment. Additionally, there are a few new monitoring tools to
get the status and monitor an LPAR activity in an AIX virtualization environment

A new AIX command, lparstat, is available to get information about the LPAR
profile and current use. Example 5-1 shows the output of the lparstat -i
command, which prints exhaustive LPAR information, including LPAR type,
virtual processors, entitled capacity, and memory.

Example 5-1 LPAR information using lparstat command

# lparstat -i
Node Name : el9-93-11
Partition Name : MOBILE-LPAR
Partition Number : 3
Type : Shared-SMT
Mode : Capped
Entitled Capacity : 1.00
Partition Group-ID : 32771
Shared Pool ID : 0
Online Virtual CPUs : 2
Maximum Virtual CPUs : 16
Minimum Virtual CPUs : 1
Online Memory : 40960 MB
Maximum Memory : 56320 MB
Minimum Memory : 1024 MB
Variable Capacity Weight : 0
Minimum Capacity : 0.10
Maximum Capacity : 8.00
Capacity Increment : 0.01
Maximum Physical CPUs in system : 8
Active Physical CPUs in system : 8
Active CPUs in Pool : 8
Shared Physical CPUs in system : 8
Maximum Capacity of Pool : 800
Entitled Capacity of Pool : 180
Unallocated Capacity : 0.00
238 Best Practices for DB2 on AIX 6.1 for POWER Systems
Physical CPU Percentage : 50.00%
Unallocated Weight : 0
Memory Mode : Dedicated
Total I/O Memory Entitlement : -
Variable Memory Capacity Weight : -
Memory Pool ID : -
Physical Memory in the Pool : -
Hypervisor Page Size : -
Unallocated Variable Memory Capacity Weight: -
Unallocated I/O Memory entitlement : -
Memory Group ID of LPAR : -
Desired Virtual CPUs : 2
Desired Memory : 40960 MB
Desired Variable Capacity Weight : 0
Desired Capacity : 1.00
Target Memory Expansion Factor : -
Target Memory Expansion Size : -

Another use of the lparstat command is similar to vmstat. lparstat can be

used to get current rolling LPAR use and other performance counters.
Additionally, lparstat offers many command options to monitor various
performance counters. See Example 5-2, Refer to the AIX Commands
Reference at the following Web page for more information:
http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/
com.ibm.aix.doc/doc/base/commandsreference.htm

Example 5-2 lparstat command

# lparstat 1

System configuration: type=Shared mode=Capped smt=On lcpu=4

mem=40960MB psize=8 ent=1.00

%user %sys %wait %idle physc %entc lbusy vcsw phint %nsp
----- ----- ------ ------ ----- ----- ------ ----- ----- ----- 0.1 0.3
0.0 99.6 0.01 0.8 0.0 472 0 99
0.0 0.2 0.0 99.8 0.01 0.6 0.0 471 0 99
0.0 0.7 0.0 99.3 0.01 1.1 0.0 468 0 99
0.0 0.2 0.0 99.8 0.01 0.6 0.0 469 0 99
0.0 0.2 0.0 99.8 0.01 0.6 0.0 469 0 99
0.2 0.5 0.0 99.3 0.01 1.1 0.0 474 0 99
Chapter 5. LPAR considerations 239
The AIX vmstat command shows additional information for the shared processor
LPAR. As shown in Example 5-3, the last two columns are displayed only for
shared processor partition. pc is number of physical processors consumed, and
ec is the percentage of entitled capacity consumed. It can be more than 100 if the
partition is uncapped, and it is currently using more than its entitlement.

Example 5-3 vmstat for shared processor LPAR

# vmstat 1

System configuration: lcpu=4 mem=40960MB ent=1.00

kthr memory page faults cpu

----- ----------- ------------------------ ------------
-----------------------
r b avm fre re pi po fr sr cy in sy cs us sy id wa
pc ec
0 0 1062519 7866593 0 0 0 0 0 0 9 7652 358 3 3 94 0
0.06 5.9
0 0 1062520 7866592 0 0 0 0 0 0 2 62 297 0 1 99 0
0.01 1.2
0 0 1062520 7866592 0 0 0 0 0 0 3 58 277 0 0 99 0
0.01 0.7
0 0 1062520 7866592 0 0 0 0 0 0 3 918 284 0 1 99 0
0.01 1.2
0 0 1062520 7866592 0 0 0 0 0 0 5 58 264 0 0 99 0
0.01 0.7
0 0 1062520 7866592 0 0 0 0 0 0 3 59 279 0 0 99 0
0.01 0.6
0 0 1062520 7866592 0 0 0 0 0 0 3 75 294 0 0 99 0
0.01 0.6
0 0 1062520 7866592 0 0 0 0 0 0 4 81 285 0 0 99 0
0.01 0.6
0 0 1062520 7866592 0 0 0 0 0 0 1 62 289 0 0 99 0
0.01 0.6
0 0 1062520 7866592 0 0 0 0 0 0 2 64 281 0 0 99 0
0.01 0.6
0 0 1062520 7866592 0 0 0 0 0 0 11 7602 359 4 4 92 0
0.09 8.8
0 0 1062520 7866592 0 0 0 0 0 0 2 60 296 0 0 99 0

0.01 0.7
240 Best Practices for DB2 on AIX 6.1 for POWER Systems
 6

Chapter 6. Virtual I/O

DB2 performance is dependent on the I/O subsystem performance. To attain the
best possible I/O throughput, the data layouts of database tables demand special
attention from database administrators and system administrators. The chosen
I/O type has great impact on the manageability and extensibility of the DB2
storage requirements. Thus, it is critical to consider workload priorities and to
examine trade-offs between disk I/O types. You can choose between locally
attached I/O or virtual IO, or both within a partition.

In this chapter we discuss the advantages of using virtual I/O, VIOS (which is the
backbone for enabling virtual I/O), high availability for virtual I/O and best
practices for configuring for virtual I/O for a DB2 workload.

This chapter has the following sections:

› “Virtual I/O benefits” on page 243
› “VIOS features” on page 244
› “VIOS resilience” on page 250
› “VIOS sizing” on page 257
› “VIOS best practices for DB2” on page 261
© Copyright IBM Corp. 2010. All rights reserved. 241
Note: Some of the information in this chapter has been consolidated from the
following materials. For more information, you can refer to these sources:
› IBM Systems Hardware Information Center:
http://publib.boulder.ibm.com/infocenter/powersys/v3r1m5/index.js
p?topic=/iphb1/iphb1_vios_virtualioserveroverview.htm
› IBM System p Advanced POWER Virtualization Best Practices:
http://www.redbooks.ibm.com/redpapers/pdfs/redp4194.pdf
› PowerVM Virtualization on IBM System p: Introduction and Configuration:
http://www.redbooks.ibm.com/redbooks/pdfs/sg247590.pdf
› DB2 Best Practices - Improving Data Server Utilization and Management
through Virtualization:
http://download.boulder.ibm.com/ibmdl/pub/software/dw/dm/db2/best
practices/DB2BP_Virtualization_0508I.pdf
› Introduction to the VIOS Whitepaper:
http://www-03.ibm.com/systems/p/hardware/whitepapers/virtual_io.h tml
242 Best Practices for DB2 on AIX 6.1 for POWER Systems
6.1 Virtual I/O benefits
This Power VM feature provides the ability to create multiple LPARs in a system.
Each LPAR needs storage for a root disk (AIX rootvg volume group) to install the
operating system and network adapter. Without the use of virtual I/O, each LPAR
needs a physical I/O adapter to provide access to the needed storage and
network adapter. Such requirements can limit the number of LPARs in a system
to the number of available I/O adapters. However, the use of virtual I/O eliminates
such requirements, and allows creating up to 255 LPARs in a system.

Use of virtual I/O requires an LPAR with a special operating system. This LPAR is
called a virtual I/O server (VIOS). VIOS enables LPARs to share I/O resources
such as Ethernet adapters, Fibre Channel adapters, and disk storage. When an
LPAR (an LPAR running DB2, for example) is using virtual I/O, all I/O requests
are passed down to the VIOS, where VIOS performs the actual I/O operation and
returns data, in a few cases, directly to a client partition (no double buffering in
the case of Virtual SCSI).

Using the VIOS facilitates the following functions:

› Sharing of physical resources between logical partitions on the system
› Sharing of disk space
› Creating logical partitions without requiring additional physical I/O resources
› Creating more logical partitions than there are I/O slots or physical devices
available with the ability for logical partitions to have dedicated I/O, virtual I/O,
or both
› Maximizing use of physical resources on the system
› Helping to reduce the Storage Area Network (SAN) infrastructure
› Server consolidation

In most cases, DB2 workloads fluctuate over time. The fluctuation leads to
variations in the use of I/O infrastructure. Typically, in the absence of virtual I/O,
each partition is configured with sizing based of the peak I/O capacity, resulting
in under used I/O infrastructure. Sharing resources where the peaks and valleys
are complementary (for example, daytime OLTP workload with nightly batch or
reporting workloads) can result in a significant reduction in the total amount of
physical resource requirements to handle the aggregated workload for all the
partitions that have these complementary peaks and valleys. By reducing the
number of physical disk drives and networking infrastructure, virtual I/O simplifies
the management of these entities.
Chapter 6. Virtual I/O 243
 To improve overall use and performance, an objective of the DB2 sizing and
tuning exercise is to reduce or eliminate I/O bottleneck in the DB2 environment.
To that end, to attain the best possible I/O bandwidth, DB2 storage requirements
are typically sized in terms of number of disk spindles, rather than the amount of
data storage. This results in an excessive amount of unused disk space. Using
virtual I/O, such disk space can be shared for complimentary or staggered
workloads.

6.2 VIOS features

The following list details a few of the key features available by using VIOS:
› Storage virtualization
– virtual SCSI
– virtual Fibre Channel (NPIV)
› Network virtualization
› Live Partition Mobility
› Integrated Virtualization Manager
› Shared memory paging for Active Memory Sharing

6.2.1 Storage virtualization

VIOS allows virtualization of physical storage resources. The storage devices are
accessed by the client partitions either through virtual SCSI or through virtual
Fibre Channel.

Virtual SCSI
Physical adapters with attached disks or optical devices on the VIOS logical
partition can be shared by one or more client logical partitions. The VIOS offers a
local storage subsystem that provides standard SCSI-compliant logical unit
numbers (LUNs). VIOS can export a pool of heterogeneous physical storage as a
homogeneous pool of block storage in the form of SCSI disks for client LPAR to
use.

Virtual SCSI is based on a client-server paradigm. VIOS owns the physical

resources as well as the virtual SCSI adapter, and acts as a server, or SCSI
target device. The client logical partitions have a SCSI initiator referred to as the
virtual SCSI client adapter, and access the virtual SCSI targets as standard SCSI
LUNs. Configure the virtual adapters by using the HMC or Integrated
Virtualization Manager. The configuration and provisioning of virtual disk
244 Best Practices for DB2 on AIX 6.1 for POWER Systems
resources is performed by using the VIOS. Physical disks owned by the VIOS
can be either exported and assigned to a client logical partition as a whole or can
be partitioned into parts, such as logical volumes or files. The logical volumes
and files can be assigned to logical partitions. Therefore, using virtual SCSI, you
can share adapters as well as disk devices. To make a physical volume, logical
volume, or file available to a client logical partition requires that it be assigned to
a virtual SCSI server adapter on the VIOS. The client logical partition accesses
its assigned disks through a virtual-SCSI client adapter. The virtual-SCSI client
adapter recognizes standard SCSI devices and LUNs through this virtual
adapter.

Figure 6-1 shows a standard virtual SCSI configuration.

Figure 6-1 Virtual SCSI configuration

After virtual devices are configured, those devices appear as regular hdisks in
AIX device configuration. The hdisk can be used like an ordinary hdisk for
creating a logical volume group. The important point is that from DB2’s
perspective, DB2 is unaware of the fact that it is using virtual storage. Therefore,
all of the DB2 storage management aspects, such as including containers, table
spaces, automatic storage remain as is, even for the virtual devices
Chapter 6. Virtual I/O 245
Virtual Fibre Channel (NPIV)
With N_Port ID Virtualization (NPIV), you can configure the managed system so
that multiple logical partitions can access independent physical storage through
the same physical Fibre Channel adapter.

To access physical storage in a typical SAN that uses Fibre Channel, the physical
storage is mapped to LUNs. The LUNs are mapped to the ports of physical Fibre
Channel adapters. Each physical port on each physical Fibre Channel adapter is
identified using one worldwide port name (WWPN).

NPIV is a standard technology for Fibre Channel networks that enables you to
connect multiple logical partitions to one physical port of a physical Fibre
Channel adapter. Each logical partition is identified by a unique WWPN, which
means that you can connect each logical partition to independent physical
storage on a SAN.

NPIV support is included with PowerVM Express, Standard, and Enterprise

Edition and supports AIX V5.3 and AIX V6.1 partitions on selected POWER6
processor-based servers Power 520, 550, 560, and 570, with an 8 GB Fibre
Channel Host Bus Adapter (HBA).

To enable NPIV on the managed system, you must create a VIOS logical
partition (version 2.1 or later) that provides virtual resources to client logical
partitions. You assign the physical Fibre Channel adapters (that support NPIV) to
the VIOS logical partition. Then you connect virtual Fibre Channel adapters on
the client logical partitions to virtual Fibre Channel adapters on the VIOS logical
partition. A virtual Fibre Channel adapter is a virtual adapter that provides client
logical partitions with a Fibre Channel connection to a SAN through the VIOS
logical partition. The VIOS logical partition provides the connection between the
virtual Fibre Channel adapters on the VIOS logical partition and the physical
Fibre Channel adapters on the managed system.

Figure 6-2 on page 247 shows a managed system configured to use NPIV.
246 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 6-2 a managed system configured to use NPIV

Figure 6-2 shows the following connections:

› A storage area network (SAN) connects three units of physical storage to a
physical Fibre Channel adapter that is located on the managed system. The
physical Fibre Channel adapter is assigned to the VIOS and supports NPIV.
› The physical Fibre Channel adapter connects to three virtual Fibre Channel
adapters on the VIOS. All three virtual Fibre Channel adapters on the VIOS
connect to the same physical port on the physical Fibre Channel adapter.
› Each virtual Fibre Channel adapter on the VIOS connects to one virtual Fibre
Channel adapter on a client logical partition. Each virtual Fibre Channel
adapter on each client logical partition receives a pair of unique WWPNs. The
client logical partition uses one WWPN to log into the SAN at any given time.
The other WWPN is used when you move the client logical partition to
another managed system.
Chapter 6. Virtual I/O 247
 Using their unique WWPNs and the virtual Fibre Channel connections to the
physical Fibre Channel adapter, the operating systems that run in the client
logical partitions discover, instantiate, and manage their physical storage located
on the SAN. In Figure 6-2 on page 247, Client logical partition 1 accesses
Physical storage 1, Client logical partition 2 accesses Physical storage 2, and
Client logical partition 3 accesses Physical storage 3. The VIOS cannot access
and does not emulate the physical storage to which the client logical partitions
have access. The VIOS provides the client logical partitions with a connection to
the physical Fibre Channel adapters on the managed system.

There is always a one-to-one relationship between virtual Fibre Channel

adapters on the client logical partitions and the virtual Fibre Channel adapters on
the VIOS logical partition. That is, each virtual Fibre Channel adapter on a client
logical partition must connect to only one virtual Fibre Channel adapter on the
VIOS logical partition, and each virtual Fibre Channel on the VIOS logical
partition must connect to only one virtual Fibre Channel adapter on a client
logical partition.

Using SAN tools, you can zone and mask LUNs that include WWPNs that are
assigned to virtual Fibre Channel adapters on client logical partitions. The SAN
uses WWPNs that are assigned to virtual Fibre Channel adapters on client
logical partitions the same way it uses WWPNs that are assigned to physical
ports.

Note: DB2 is not aware of the fact that it is using virtual storage. All of the
storage management best practices and methods discussed so far are equally
applied to virtual storage as well.

6.2.2 Network virtualization

The hypervisor on a POWER 5 and POWER 6 system provide virtual network
support. Without requiring additional hardware or external cables, a virtual LAN
(VLAN) facilitates high-speed virtual Ethernet communication paths among
multiple partitions within a physical system that run AIX, Linux, and other
operating systems. You can dynamically create virtual Ethernet segments and
restrict access to a VLAN segment to meet security or traffic segregation
requirements. A virtual Ethernet has the same characteristics as a
high-bandwidth, physical Ethernet network and supports multiple networking
protocols, such as IPv4, IPv6, and ICMP.

VIOS provides the virtual networking technologies discussed in the next three
sections.
248 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Shared Ethernet adapter (SEA)
The shared Ethernet adapter (SEA) hosted in the VIOS acts as a layer-2 bridge
between the internal virtual and external physical networks. The SEA enables
partitions to communicate outside the system without having to dedicate a
physical I/O slot and a physical network adapter to a client partition

SEA failover
SEA failover provides redundancy by configuring a backup SEA on a VIOS
logical partition that can be used if the primary SEA fails. The network
connectivity in the client logical partitions continues without disruption.

Link aggregation (or EtherChannel)

A Link Aggregation (or EtherChannel) device is a network port-aggregation
technology that allows several Ethernet adapters to be aggregated. The adapters
can then act as a single Ethernet device. Link Aggregation helps provide more
throughput over a single IP address than is possible with a single Ethernet
adapter.

6.2.3 Live Partition Mobility

Live Partition Mobility provides the ability to move AIX and Linux logical partitions
from one system to another. The mobility process transfers the system
environment, including the processor state, memory, attached virtual devices,
and connected users.

Active Partition Mobility allows you to move AIX and Linux logical partitions that
are running, including the operating system and applications, from one system to
another. The logical partition and the applications running on that migrated
logical partition do not need to be shut down.

Inactive Partition Mobility allows you to move a powered off AIX and Linux logical
partition from one system to another.

The VIOS plays a main role in live partition mobility. The Live Partition Mobility is
initiated from the HMC. The HMC communicates with the VIOS on the two
systems to initiate the transfer. The VIO servers on both system talk to each
other, and transfer the system environment including the processor state,
memory, attached virtual devices, and connected users.

For more information about Live Partition Mobility you can refer to Chapter 7,
“Live Partition Mobility” on page 269.
Chapter 6. Virtual I/O 249
6.2.4 Integrated Virtualization Manager
The Integrated Virtualization Manager (IVM) provides a Web-based system
management interface and a command-line interface that you can use to
manage IBM Power Systems servers and IBM BladeCenter blade servers that
use the IBM VIOS. On the managed system, you can create logical partitions,
manage virtual storage and virtual Ethernet, and view service information related
to the server. The IVM is included with the VIOS, but it is available and usable
only on certain platforms, and where no Hardware Management Console (HMC)
is present.

If you install the VIOS on a supported server, and if there is no HMC attached to
the server when you install the VIOS, then the IVM is enabled on that server. You
can use the IVM to configure the managed system through the VIOS.

6.2.5 Shared memory paging for Active Memory Sharing

Logical partitions can share the memory in the shared memory pool by using the
PowerVM Active Memory Sharing technology (or shared memory). Instead of
assigning a dedicated amount of physical memory to each logical partition that
uses shared memory (hereafter referred to as shared memory partitions), the
hypervisor constantly provides the physical memory from the shared memory
pool to the shared memory partitions as needed. The hypervisor provides
portions of the shared memory pool that are not currently being used by shared
memory partitions to other shared memory partitions that need to use the
memory. When a shared memory partition needs more memory than the current
amount of unused memory in the shared memory pool, the hypervisor stores a
portion of the memory that belongs to the shared memory partition in auxiliary
storage. Access to the auxiliary storage is provided by a VIOS logical partition.
When the operating system attempts to access data that is located in the
auxiliary storage, the hypervisor directs a VIOS to retrieve the data from the
auxiliary storage and write it to the shared memory pool so that the operating
system can access the data.

6.3 VIOS resilience

Data storage subsystem performance and high availability is crucial for DB2
workloads. The virtual I/O must continue offering the same level of resiliency as
direct attached storage. By design, VIOS is extremely robust. It primarily runs
device drivers and does not run any application workloads.
250 Best Practices for DB2 on AIX 6.1 for POWER Systems
 To provide high availability of virtual IO to the client partitions, redundancy can be
built into VIOS itself by using a combination of the following items:
› Redundant physical hardware
› Network Interface Backup Ethernet configuration
› SEA failover configuration
› Storage Logical Volume Manager (LVM) mirroring and RAID configurations
› SAN multipath I/O (MPIO)
› RAID protected storage (RAID provided either by the storage subsystem or by
a RAID adapter)
› Hot-pluggable network adapters instead of built-in integrated network
adapters

Two or more VIOSs and redundant devices can provide improved software
maintenance and hardware replacement strategies.

6.3.1 Dual VIOS

With multiple virtual I/O client partitions, running DB2, dependent on a VIOS for
I/O demands, you can implement dual VIOSs, duplicate paths and devices to
provide additional system service and configuration options.

Fundamentally, the primary reasons for redundant VIOSs configuration include:

› Future hardware expansion and new function
› Unscheduled outages due to human intervention
› Unscheduled outages due to physical device failure or natural events
› Scheduled outages required for VIOS maintenance
› Isolation of network and storage workload to provide increased virtual I/O
client partition performance
› Multiple multipath codes such as MPIO and device redundancy

The following sections show the storage and network redundancy options
provided by a dual VIOS environment.
Chapter 6. Virtual I/O 251
Virtual SCSI redundancy
Virtual SCSI redundancy can be achieved using MPIO at client partition and
VIOS level. Figure 6-3 displays a setup using MPIO in the client partition.

Client Partition

hdiskx

MPIO

vSCSI vSCSI
Client Adapter Client Adapter

I/O Server 1 I/O Server 2

Figure 6-3 MPIO attributes

Two VIOSs host disks for a client partition. The client is using MPIO to access a
SAN disk. From the client perspective, the following situations can be handled
without causing downtime for the client:
› Either path to the SAN disk can fail, but the client is still able to access the
data on the SAN disk through the other path. No action has to be taken to
reintegrate the failed path to the SAN disk after repair if MPIO is configured
› Either VIOS can be rebooted for maintenance. This results in a temporary
simultaneous failure of one path to the SAN disk, as described before.

Note: It is recommend to use a minimum of two Fibre Channel adapters in

each VIOS for adapter redundancy.
252 Best Practices for DB2 on AIX 6.1 for POWER Systems
The MPIO for virtual SCSI devices only supports failover mode. For any given
virtual SCSI disk, a client partition uses a primary path to one VIOS and fail over
to the secondary path to use the other VIOS. Only one path is used at a given
time even when both paths are enabled.

To balance the load of multiple client partitions across two VIOSs, the priority on
each virtual SCSI disk on the client partition can be set to select the primary path
and, therefore, a specific VIOS. The priority is set on a per virtual SCSI disk basis
using the chpath command as shown in the following example (1 is the highest
priority):
chpath -l hdisk0 -p vscsi0 -a priority=2

Due to this granularity, a system administrator can specify whether all the disks
or alternate disks on a client partition use one of the VIOSs as the primary path.
The recommended method is to divide the client partitions between the two
VIOSs.

For detailed information about the various device settings mentioned in

Figure 6-3 on page 252, refer to IBM Redbooks publication IBM PowerVM
Virtualization Managing and Monitoring, SG24-7590.

Virtual network redundancy

The two common methods available to provide virtual I/O client partition network
redundancy in dual VIOSs configurations are:
› Network Interface Backup (NIB)
› Shared Ethernet Adapter (SEA) Failover
Chapter 6. Virtual I/O 253
Network Interface Backup
Figure 6-4 shows a highly-available network configuration using dual VIOSs.

Figure 6-4 Network Interface Backup using dual VIOSs

This configuration uses virtual Ethernet adapters created using default virtual
LAN IDs with the physical Ethernet switch using untagged ports only. In this
example, VIOS 1 has a SEA that provides external connectivity to client partition
1 through the virtual Ethernet adapter using virtual LAN ID 2. VIOS 2 also has a
SEA that provides external connectivity to the client partition through the virtual
Ethernet adapter using virtual LAN ID 3. Client partition 2 has a similar setup,
except that the virtual Ethernet adapter using virtual LAN ID 2 is the primary and
virtual LAN ID 3 is the backup. This enables client partition 2 to get its primary
connectivity through VIOS 1 and backup connectivity through VIOS 2.
254 Best Practices for DB2 on AIX 6.1 for POWER Systems
Client partition 1 has the virtual Ethernet adapters configured using Network
Interface Backup such that the virtual LAN ID 3 network is the primary and virtual
LAN ID 2 network is the backup, with the IP address of the default gateway to be
used for heartbeats. This enables client partition 1 to get its primary connectivity
through VIOS 2 and backup connectivity through VIOS 1.

If the primary VIOS for an adapter becomes unavailable, the Network Interface
Backup mechanism detects this because the path to the gateway is broken. The
Network Interface Backup setup fails over to the backup adapter that has
connectivity through the backup VIOS.

SEA failover
SEA failover is implemented on the VIOS using a bridging (layer-2) approach to
access external networks. SEA failover supports IEEE 802.1Q VLAN-tagging,
unlike Network Interface Backup.

With SEA failover, two VIOSs have the bridging function of the SEA to
automatically fail over if one VIOS is unavailable or the SEA is unable to access
the external network through its physical Ethernet adapter. A manual failover can
also be triggered.
Chapter 6. Virtual I/O 255
As shown in Figure 6-5, both VIOSs attach to the same virtual and physical
Ethernet networks and VLANs, and both virtual Ethernet adapters of both
Shared Ethernet Adapters have the access external network flag enabled. An
additional virtual Ethernet connection must be set up as a separate VLAN
between the two VIOSs and must be attached to the Shared Ethernet Adapter
(SEA) as a control channel. This VLAN serves as a channel for the exchange of
keep-alive or heartbeat messages between the two VIOSs that controls the
failover of the bridging functionality. No network interfaces have to be attached to
the control channel Ethernet adapters. The control channel adapter is dedicated
and on a dedicated VLAN that is not used for any other purpose.

Figure 6-5 SEA failover using dual VIOSs

In addition, the SEA in each VIOS must be configured with differing priority
values. The priority value defines which of the two SEAs is the primary (active)
and which is the backup (standby). The lower the priority value, the higher the
priority (for example, priority=1 is the highest priority).
256 Best Practices for DB2 on AIX 6.1 for POWER Systems
 You can also configure the SEA with an IP address that it periodically pings to
confirm that network connectivity is available. This is similar to the IP address to
ping that can be configured with Network Interface Backup (NIB). If you use NIB,
you have to configure the reachability ping on every client compared to doing it
when on the SEA.

It is possible that during an SEA failover, the network drops up to 15–30 packets
while the network reroutes the traffic.

6.4 VIOS sizing

The only certain way to size any server is to run it with the real workload, monitor,
and then tune.

The following sections provide a starting point for tuning.

› The IBM Systems Hardware Information Center has detailed planning
calculations to help with CPU and memory planning:
http://publib.boulder.ibm.com/infocenter/eserver/v1r3s/topic/iphb1/i
phb1_vios_planning.htm
› To use the IBM Systems Workload Estimator tool to help size your server,
see:
http://www-912.ibm.com/wle/EstimatorServlet

The following guidelines are intended to get you started and might need
additional adjustments after they are run and monitored in production, but they
enable you to accurately plan how much CPU and memory to use.

6.4.1 VIOS memory sizing

The VIOS, similar to other workloads, requires system memory to operate. The
memory requirement for servicing virtual I/O is insignificant. The memory
requirement for the VIOS mainly depends on network requirements. The memory
used for network communication is used to buffer network traffic, so detailed
planning involves knowing such factors as the messages sizes (MTU or jumbo
frames) and how many SEAs to use

The easiest rule to follow is that if you have a simple VIOS that is only bridging a
couple of networks and never uses jumbo frames or a larger MTU, 512 MB is
needed by the VIOS. System configurations with 20–30 clients, many LUNs, and
a generous vhost configurations might need increased memory to suit
performance expectations.
Chapter 6. Virtual I/O 257
 If there is a possibility that you might start bridging more networks and use jumbo
frames, use 1 GB of RAM for the VIOS. If you have enough memory to spare, the
best situation is to use 1 GB for the VIOS to allow the maximum flexibility. When
10 Gbps Ethernet adapters are commonly implemented, this memory
requirement might require revision. If VIOS is using IVE/HEA as an SEA, the
LPAR might need more than 1 GB.

6.4.2 VIOS CPU sizing

The CPU planning has two major parts:
› The CPU required to support the virtual network
› The CPU required to support the virtual disk

CPU requirements to support the virtual network

When client partitions with virtual Ethernets need to access the external network,
they need to bridge through the VIOS using the SEA so the VIOS sees all the
network traffic flowing between the external network and the client partitions.

The CPU requirements on the VIOS for the network depends on the actual
network traffic that runs through the VIOS and not on the actual number of
network adapters installed on it. Minimal CPU is needed to support a network
adapter card. For every network packet, however, the CPU has to calculate
things such as the network checksum

Table 6-1 lists CPU requirements based on POWER5 system to push 1 GB

network traffic. We can use this has a starting point.

Table 6-1 Approximate CPU amount VIOS needs for 1 GB of network traffic
MTU (bytes) 1500 9000 or jumbo frames

CPU speed (GHz)

1.5 1.1 0.55

1.65 1.0 0.5

1.9 0.87 0.44

2.2 0.75 0.38

Let us see an example. Suppose that you have a VIOS with 4 gigabit network
adapters and you know that during normal operations you have about 100 Mb of
traffic overall. However, you also know that during a backup window you use a full
gigabit of network traffic for two hours at night.
258 Best Practices for DB2 on AIX 6.1 for POWER Systems
These values can be translated into a CPU requirement to support the network.
This can best be done by using the shared CPU model. If we assume that your
server has 1.65 GHz CPUs and you are using an MTU of 1500, we can calculate
that during normal operation you only need 0.1 of a CPU. During peak loads, you
need 1.0 CPU. If we assume that your user base is not using the system at night
(thus the backup), there is plenty of unused CPU in the free pool that can be
used for the CPU requirement here. You can configure the VIOS partition as a
shared uncapped partition with 0.1 entitlement with 1 virtual CPU. This
guarantees 0.1 of a CPU to sustain the daily network usage, but by using the
uncapped CPU resources, we can allow the VIOS to grow to 1.0 CPU if required
using spare CPU cycles from the CPU pool.

Remember that you need CPU to support network traffic and that adding
additional network cards (providing the same network traffic) does not require
additional CPU. Using Table 6-1 on page 258, estimate a value for the required
network bandwidth to support normal operations. Guarantee this amount (plus
the disk value from the following section) to the logical partition as the processing
units. Use uncapped CPUs on the servers profile to allow the VIOS to use spare
processing from the free pool to handle any spikes that might occur.

CPU requirements to support the virtual disk

The disk CPU requirement is more difficult to work out accurately because it
involves knowing detailed information about your I/O, such as block sizes and
number of I/Os per second. If you know this information, the IBM Systems
Hardware Information Center has the detailed planning calculations.

For a rough guideline, it is probably easier to work out what the disks you are
using can provide and make an estimate as to how busy you think these disks
will be. For example, consider a simple VIOS that has a basic internal set of four
SCSI disks. These disks are used to provide all of the I/O for the clients and are
10 K RPM disks. We use a typical workload of 8 KB blocks. For these disks, a
typical maximum is around 150 I/Os per second, so this works out to be about
0.02 of a CPU. A small amount.
Chapter 6. Virtual I/O 259
If we plot the amount of CPU required against the number of I/Os per second for
the 1.65 GHz CPUs for the various I/O block sizes, we get a the data shown in
Figure 6-6.

Figure 6-6 Estimated size of storage array to drive I/O versus VIOS CPU

These I/O numbers and the storage subsystem sizing assume that the storage
subsystem is being driven at 100% by the virtual I/O clients only (so every I/O is
virtual disk) and the disk subsystems have only been placed on the graph to
indicate the performance you need to generate this sort of load.

It is important to remember that we are not saying the bigger the disk subsystem,
the more CPU power you need. What we actually find is that we need the most
powerful storage subsystem offered to require any significant amount of CPU in
the VIOS from a disk usage perspective. As an example, a mid-range disk
subsystem running at full speed is in the region of the yellow/green boundary
(mid-range to high end) on the graph when configured with more than 250 drives.
In most cases, this storage is not 100% dedicated to the VIOS. However, if this
was the case, even when running at full speed, a Virtual I/O only needs one CPU
for most block sizes.

Most I/O requests for systems are in the lower left section, so a starting figure of
0.1 to 0.2 CPU to cater for a disk is a good starting point in most cases.
260 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Note: We always suggest testing a configuration before putting it into
production.

6.4.3 Sizing dual VIOSs

If you are planning a system with dual VIOSs, you can size each VIOS to support
the workload of one half of the clients. Use uncapped CPU to provide enough
reserve CPU in case a VIOS is removed from the configuration for service.

6.4.4 Sizing VIOS for Live Partition Mobility

When configuring a VIOS to do an active Live Partition Mobility, consider the
additional CPU that is required on the source and destination VIOS to do the
migration. As discussed in Chapter 7, “Live Partition Mobility” on page 269,
during an active Live Partition Mobility, the source VIOS transfers the client
partitions configuration and memory to the destination VIOS over the network.

Because network traffic requires additional CPU, we need to size both VIOS
adequately to minimize the migration time. The actual amount of CPU required
varies depending upon network parameters and the processor used on the
system. Therefore, it is recommended to configure both the VIOSs as shared
uncapped partitions and allocate one additional VCPU on top of what is required
for processing the required virtual I/O and virtual network requests. This allows
the VIOSs to make use of available CPU from the free pool during the migration
and reduces the total time required for doing the migration.

6.5 VIOS best practices for DB2

This section discusses the practices that are suggested in configuring the VIOS
for servicing a client partition that runs a DB2 workload.

6.5.1 Multi-path I/O

Because I/O is important to a DB2 workload, we need to provide both
redundancy and efficient bandwidth to do the I/O. This can be achieved by using
multi-path I/O capability on the VIOS.
Chapter 6. Virtual I/O 261
 For DB2 DataWarehouse workloads, bandwidth might be the key requirement,
whereas for DB2 OLTP workloads, the total number of I/Os that can be
processed might be the key requirement. MPIO address both the requirements
by making use of multiple adapters and providing multiple paths,

Multipathing for the physical storage in the VIOS provides failover physical path
redundancy and load-balancing. The multipathing solutions available in the VIOS
include MPIO, as well as solutions provided by the storage vendors.

The multipath I/O software needs to be configured only on the VIOS. There is no
special configuration that needs to be done on the virtual I/O client partitions to
make use of the multipath I/O at the VIOS.

For details on configuring multipath I/O, see the corresponding storage vendor’s
documentation.

6.5.2 Networking
Because a DB2 workload depends on network to talk to the clients, we need to
provide both redundancy as well as efficient bandwidth for the network traffic too.
This can be achieved by configuring link aggregation for the SEA adapter on the
VIOS.

Link aggregation is a network port aggregation technology that allows several

Ethernet adapters to be aggregated together to form a single pseudo-Ethernet
adapter. This technology is often used to overcome the bandwidth limitation of a
single network adapter and avoid bottlenecks when sharing one network adapter
among many client partitions. For more information about the types of link
aggregation technologies, refer to PowerVM Virtualization on IBM System p:
Introduction and Configuration Fourth Edition, SG24-7940.
262 Best Practices for DB2 on AIX 6.1 for POWER Systems
To provide redundancy in case of port or switch failure, for single VIOS
configuration, we can further configure a NIB on the SEA adapter, as shown in
Figure 6-7.

Figure 6-7 Link aggregation using a single VIOS

For dual VIOS configurations, redundancy can be provided by using either SEA
failover or configuring a NIB on the client partition as discussed in“Virtual
network redundancy” on page 253.
Chapter 6. Virtual I/O 263
6.5.3 CPU settings for VIOS
If the I/O response time is not important for the DB2 workload running on the
client partition, and if the CPU sizing requirement for the VIOS is less than 1.0
CPU entitlement, we can leave the CPU entitlement less than 1.0, but make it as
a shared uncapped partition. This allows it to make use of available processing
power in the free pool when required during peak load.

On the other hand, if the response time for the I/O is important for the DB2
workload, we need to make sure that the VIOS can get access to a CPU and
service the I/O request. This can be done by either configuring the VIOS with a
whole CPU entitlement (1.0) and making it a shared uncapped partition or, on a
POWER6 system, we can configure it as Dedicate-Donate partition. Both these
methods guarantee that the VIOS can immediately service a I/O request. Even
though we are allocating more CPU entitlement than required for the VIOS,
because we have configured it as shared or dedicate-donate, other partitions can
make use of the unused capacity. We need to make sure the other LPARs are
configured as shared-uncapped partitions with appropriate VCPUs to make use
of the available processing capacity.

6.5.4 SCSI queue depth

Typically a DB2 workload makes use of RAID storage with multiple physical disks
configured for each LUN. The default SCSI command queue depth of these
LUNs, as seen by the VIOS, might not be sufficient. Increasing them might
provide improved performance depending on the configuration. A general rule of
thumb is to configure queue depth to eight times the number of physical disks.

Use the following command to change the queue depth on the VIOS:
chdev -dev hdiskN -attr queue_depth=x

Note: Changing the queue depth of the physical LUN might require changing
the queue depth for the virtual disk on the client partition to achieve optimal
performance.

There are several other factors to be taken into consideration when changing the
queue depth. These factors include the value of the queue_depth attribute for all
of the physical storage devices on the VIOS being used as a virtual target device
by the disk instance on the client partition. It also includes the maximum transfer
size for the virtual SCSI client adapter instance that is the parent device for the
disk instance.

The maximum transfer size for virtual SCSI client adapters is set by the VIOS,
which determines that value on the resources available on that server and the
264 Best Practices for DB2 on AIX 6.1 for POWER Systems
maximum transfer size for the physical storage devices on that server. Other
factors include the queue depth and maximum transfer size of other devices
involved in mirrored volume group or MPIO configurations. Increasing the queue
depth for a few devices might reduce the resources available for other devices on
that same parent adapter and decrease throughput for those devices.

The most straightforward configuration is when there is a physical LUN used as

the virtual target device. For the virtual SCSI client device queue depth to be
used effectively, it must not be any larger than the queue depth on the physical
LUN. A larger value wastes resources without additional performance. If the
virtual target device is a logical volume, the queue depth on all disks included in
that logical volume must be considered. If the logical volume is being mirrored,
the virtual SCSI client queue depth must not be larger than the smallest queue
depth of any physical device being used in a mirror.

When mirroring, the LVM writes the data to all devices in the mirror, and does not
report a write as completed until all writes have completed. Therefore, throughput
is effectively throttled to the device with the smallest queue depth. This applies to
mirroring on the VIOS and the client.

We suggest that you have the same queue depth on the virtual disk as the
physical disk. If you have a volume group on the client that spans virtual disks,
keep the same queue depth on all the virtual disks in that volume group. This is
important if you have mirrored logical volumes in that volume group, because the
write does not complete before the data is written to the last disk.

In MPIO configurations on the client, if the primary path has a much greater
queue depth than the secondary, there might be a sudden loss of performance
as the result of a failover.

The virtual SCSI client driver allocates 512 command elements for each virtual
I/O client adapter instance. Two command elements are reserved for the adapter
to use during error recovery. Three command elements are reserved for each
device that is open to be used in error recovery. The rest are left in a common
pool for use in I/O requests. As new devices are opened, command elements are
removed from the common pool. Each I/O request requires one command
element for the time that it is active on the VIOS.

Increasing the queue depth for one virtual device reduces the number of devices
that can be open at one time on that adapter. It also reduces the number of I/O
requests that other devices can have active on the VIOS.
Chapter 6. Virtual I/O 265
6.5.5 Dual VIOS
To provide the DB2 workload running on the client partition high availability to
virtual I/O and virtual network, it is suggested to configure and use dual VIOSs
as discussed in 6.3.1, “Dual VIOS” on page 251.

6.5.6 SEA threading

The VIOS enables you to virtualize both disk and network traffic for the virtual I/O
clients. The main difference between these types of traffic is the persistence of
them. If the VIOS has to move network data around, it must do this immediately
because network data has no persistent storage. For this reason, the network
services provided by the VIOS (such as the SEA) run with the highest priority.
Disk data for virtual SCSI devices is run at a lower priority than the network
because the data is stored on the disk and there is less of a danger of losing it
due to time outs. The devices are also normally slower in speed.

The shared Ethernet process of the VIOS prior to Version 1.3 runs at the
interrupt level that was optimized for high performance. With this approach, it ran
with a higher priority than the virtual SCSI if there was high network traffic. If the
VIOS did not provide enough CPU resource for both, the virtual SCSI
performance can experience a degradation of service.

With VIOS Version 1.3, the shared Ethernet function can be implemented using
kernel threads. This enables a more even distribution of the processing power
between virtual disk and network.

This threading can be turned on and off per SEA by changing the thread attribute
and can be changed while the SEA is operating without any interruption to
service. A value of 1 indicates that threading is to be used and 0 indicates the
original interrupt method:
$ chdev -dev ent2 -attr thread=0

The performance difference without threading works out at around 8% less

(using our intensive test loads) CPU needed for the same network throughput.
With the burst nature of network traffic, however, we suggest enabling threading
(this is the default). By this, we mean that network traffic comes in spikes, as
users log on or as Web pages load, for example. These spikes might coincide
with disk access. For example, a user logs on to a system, generating a network
activity spike, because during the logon process some form of password
database stored on the disk is most likely accessed, or the user profile read.
266 Best Practices for DB2 on AIX 6.1 for POWER Systems
 The one scenario where you must consider disabling threading is where you
have a VIOS dedicated for network and another dedicated for disk. This is only
recommended when mixing extreme disk and network loads together on a CPU
constricted server.

As discussed in 6.4, “VIOS sizing” on page 257, usually the network CPU
requirements are greater than the disk. In addition, you probably have the disk
VIOS setup to provide a network backup with SEA failover if you want to remove
the other VIOS from the configuration for scheduled maintenance. In this case,
you have both disk and network running through the same VIOS, so threading is
recommended.

6.5.7 Best practices summary

THe following list details the best practices in configuring VIOS and the client
LPAR for running a DB2 workload using virtual I/O.
› For storage access redundancy and bandwidth, configure MPIO on the VIOS
to access the storage
› For network access redundancy and bandwidth, use link aggregation of
network adapters to configure the SEA adapter.
› For network access redundancy on the client LPAR, use NIB or SEA failover
with dual VIOS configuration.
› VIOS requires modest CPU to process virtual I/O request. However, if the
DB2 workload is sensitive to the I/O latency, configure the VIOS as a shared
uncapped partition with a whole CPU entitlement, or on a POWER6 system,
configure it as Dedicate-Donate partition. This results in faster scheduling of
I/O operations and therefore improved Virtual I/O performance.
› The queue depth of the virtual disks on the client LPAR and the physical disks
on the VIOS is set properly to achieve optimal performance. The default value
is often inadequate for most database configurations
› For high availability of virtual I/O and virtual network for the client LPAR, it is
recommended to configure and use dual VIOSs
› Memory required for VIOS partition is insignificant. Allocating more memory
does not result in any performance improvement
› When having a mix of both heavy virtual I/O and virtual network requirements,
it might benefit to configure a separate VIOS dedicated for the virtual network
and turning off threading on the SEA.
Chapter 6. Virtual I/O 267
268 Best Practices for DB2 on AIX 6.1 for POWER Systems
 7

Chapter 7. Live Partition Mobility

Live Partition Mobility (LPM) allows you to migrate partitions that are running AIX
and Linux operating systems and their hosted applications from one physical
server to another without disrupting the infrastructure services. The migration
operation, which takes a few seconds, maintains complete system transactional
integrity. The migration transfers the entire system environment, including
processor state, memory, attached virtual devices, and connected users. LPM
requires a strict environment. Numerous pre-requisites need to be met before
being able to migrate a partition (active or inactive migration) to another system.
Other Power technology-based processor types (Power 6 systems), PowerHA
and N-PIV attached storage are among the new technologies that are now
supported with LPM. For more information about PowerVM features, see 1.5,
“Introduction to PowerVM virtualization” on page 21.
This chapter covers LPM pre-requisites in detail. We describe the migration of a
logical partition managed by two HMCs, each one managing both servers. This
chapter is not meant to describe how to configure LPM but to describe the best
practices to configure the environment for a DB2 setup.
We cover the following topics in this chapter:
› “LPM overview” on page 270
› “LPM” on page 271
› “DB2 migration-awareness” on page 273
› “System planning” on page 274
› “Migration process and flow” on page 316
› “Mobility in action” on page 322
› “LPM: Summary of actions” on page 354
© Copyright IBM Corp. 2010. All rights reserved. 269
7.1 LPM overview
Power VM LPM allows you to migrate running and powered-off AIX partitions and
their hosted applications from one POWER6 box to another with minimal
disruptions to the applications hosted on these platforms and without shutting
down the servers.

7.1.1 Active migration

Using active migration, a running partition is moved from a source system to a
destination system with no disruption of partition operation or user service.

An active migration performs the same operations as an inactive migration

except that the operating system, the applications, and the services they provide
are not stopped during the process. The physical memory content of the logical
partition is copied from system to system allowing the transfer to be
imperceptible to users.

During an active migration, the applications continue to handle their normal

workload. Disk data transactions, running network connections, user contexts,
and the complete environment are migrated without any loss. Migration can be
activated any time on any production partition. There is no limitation on a
partition’s computing and memory configuration. Multiple migrations can be
executed concurrently. Both inactive and active migrations might involve
partitions with any processing unit and memory size configuration.

7.1.2 Inactive migration

Inactive migration moves the definition of a powered off logical partition from one
system to another, along with its network and disk configuration. No additional
change in network or disk setup is required and the partition can be activated as
soon as migration is completed.

The inactive migration procedure takes care of the re-configuration of involved

systems.
› A new partition is created on the destination system with the same
configuration present on the source system.
› Network access and disk data is preserved and made available to the new
partition.
› On the source system, the partition configuration is removed and all involved
resources are freed.
270 Best Practices for DB2 on AIX 6.1 for POWER Systems
 If a system is down due to scheduled maintenance or not in service for other
reasons, an inactive migration might be performed. It is executed in a controlled
way and with minimal administrator interaction so that it can be safely and
reliably performed in a short time frame. When the service provided by the
partition cannot be interrupted, its relocation can be performed with no loss of
service by using the active migration feature.

Extending this technology to DB2 and its server installations, we can make use of
LPM to move a database server, without shutting down the database engine,
from one physical system to a second physical system while it is servicing
transactions. LPM uses a simple and automated procedure to migrate the entire
system environment, including processor state, memory, attached virtual
devices, and connected users with no application downtime.

7.2 LPM
Infrastructure flexibility has become a key criteria when designing and deploying
information technology solutions. Imagine you have a DB2 database server
running on a hardware system that has exhausted all its resources. The server is
running 2 processors at 100% use. Response times are slow and users are
complaining.
› Move that DB2 database server from that system to a system that has
additional resources without interrupting service to the users
› Perhaps the system needs to go down for scheduled maintenance such as a
hardware upgrade or a firmware upgrade or even a component replacement?
› You need to do a new system deployment so that the workload running on an
existing system is migrated to a new, more powerful one.
› If a server indicates a potential failure, you can move its logical partitions to
another system before the failure occurs.
› You want to conserve energy by moving a partition to another system during
off-peak periods.

How can we do this?

The answer is LPM. With LPM, it is possible to meet continuously stringent

service level agreements, efficiently manage the infrastructure, and minimize the
impact to users.
Chapter 7. Live Partition Mobility 271
 Note: LPM can move an entire Logical Partition from one system to another
while it is running, with almost no impact to users. It can move the entire
LPAR, including the OS. It requires a minimum of a POWER6 system based
on the POWER6 processor, PowerVM Enterprise Edition, and all I/O must be
through the VIOS.

However, while LPM provides many benefits, it does not perform the following
tasks:
› LPM does not do automatic load balancing.
› LPM does not provide a bridge to new functions. Logical partitions must be
restarted and possibly reinstalled to take advantage of new features.
› LPM does not protect you from system failures, so it does not replace
high-availability software such as the IBM HACMP high-availability cluster
technology.

Note: To summarize, LPM is not:

› A replacement for DB2 HADR (high availability disaster recovery) or Tivoli
Systems Automation for Multiplatforms
› Automatic
› A disaster recovery solution

To use the partition migration, there are planning steps that are necessary.
Although an in-depth discussion on the technologies that supports the flexibility
is outside of the scope of this chapter, a quick checklist is provided that assures
your partition is able to migrate successfully.

Note: To summarize, LPM benefits are:

› Eliminates planned outages
› Balances workloads across systems
› Energy savings
272 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.3 DB2 migration-awareness
A migration-aware application is one that is designed to recognize and
dynamically adapt to changes in the underlying system hardware after being
moved from one system to another.

There are no changes required to make DB2 work when the partition is moved
from one system to another. DB2 has built-in autonomics and self-tuning that
enables it to adapt to changes in the underlying system after being moved from
one system to another:
› Self tuning memory manager (STMM) feature of DB2 queries the operating
system for free and available memory at regular intervals and automatically
adapts to the destination server's memory to maximize throughput.
› DB2 has the ability to change several parameters, without instance restart.
› DB2 uses OS-based Scheduler and can automatically adapt to the new CPU
resources.

Note: Most DB2 parameters are set to AUTOMATIC by default. The general
guideline is to leave these parameters AUTOMATIC.

Table 7-1 shows a list of default parameters used by DB2 version 9.7.

Table 7-1 List of default parameters (DB2 9.7) – plus many more
List of default DB2 9.7 parameters Parameter Value

Size of instance shared memory (4 KB) INSTANCE_MEMORY AUTOMATIC

Self tuning memory SELF_TUNING_MEM ON

Max storage for locK list (4 KB) LOCKLIST AUTOMATIC

Percentage of lock lists per application MAXLOCKS AUTOMATIC

Package cache size (4 KB) PCKCACHESZ AUTOMATIC

Sort heap threshold for shared sorts (4 SHEAPTHRES_SHR AUTOMATIC

KB)

Sort list heap (4 KB) SORTHEAP AUTOMATIC

Number of asynchronous page cleaners NUM_IOCLEANERS AUTOMATIC

Number of I/O servers NUM_IOSERVERS AUTOMATIC

Default prefetch size (pages) DFT_PREFETCH_SZ AUTOMATIC

Chapter 7. Live Partition Mobility 273
 List of default DB2 9.7 parameters Parameter Value

Max number of active applications MAXAPPLS AUTOMATIC

Average number of active applications AVG_APPLS AUTOMATIC

The performance characteristics (throughput, response time) of your database

workloads might change depending on the resources of the target server.

7.4 System planning

This section highlights the steps and actions to perform a Live Partition migration
of a logical partition running a DB2 database (for example, from one system to
another) without interrupting the service to the client. For an in-depth information
and installation steps, refer to the IBM Redbooks publication IBM PowerVM Live
Partition Mobility, SG24-7460.

Numerous pre-requisites need to be met prior to migration. After all

pre-requisites are met and satisfied, the HMC validates the migration. If the result
of the validation is successful, the migration can be performed either using the
GUI or the command line interface of the HMC.

In the following sections pre-requisites, configuration, and post migrations are

covered.

Note: The major part of this chapter covers systems managed by one or more
HMCs. Although this book covers systems managed by HMC, the tests on
LPM have been done with IVM-based systems. For information about IVM,
refer to the IBM Redpaper™ Integrated Virtualization Manager on IBM System
p5, REDP-4061.
274 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.4.1 Managed system’s requirements
Table 7-2 lists the managed system’s requirements.

Table 7-2 Managed system’s requirements

Managed system’s tasks Actions to undertake and remarks For details see

Managed systems type POWER6 systems and Power Blade “Managed system type” on
systems page 275

Managed system’s capabilities PowerVM APV EE “Managed system’s

capabilities” on page 276

Processor clock speed Mix of clock is supported “Processor clock speed” on

page 278

Time-of-day Synchronize time-of-day “Time-of-Day” on page 280

Managed system’s minimum EH330_046 (p595) or “Managed system’s

firmware level EM320_31 (p570) or firmware level” on page 281
EL320_040 (p520 + p550)
Available memory on destination Sufficient resources on destination “Available memory on
system destination system” on
page 281

Available processor or Sufficient resources on destination “Available processor on

processing units on destination system destination system” on
system page 281

Existing profile Mobile profile not on destination “Existing profile” on

system page 282

Managed systems’ connectivity Single or dual redundant HMC, “Managed systems’

separate HMC connectivity” on page 283

Logical Memory Block Must be identical on both systems “Logical Memory Block
(LMB)” on page 284

Battery power Supported on source system, not “Battery power” on

destination system page 284

Managed system type

Managed systems are those systems able to host a migrating partition. These
are Power architectured systems, such as Power System p6 or Power Blade.

Managed servers in an LPM must be of same type (for example source and
target servers are POWER6 servers or Power Blades). The same HMC or
redundant pair of HMCs are used with Power Systems. IVM is used for Power
Chapter 7. Live Partition Mobility 275
Blades and for Power Systems (models below p570). If you want to mix
POWER6 servers and Power blades, this is possible if the systems are managed
by IVM. Each of the servers can act as a source or a target system to the other
as long as it contains the necessary processor hardware to support it. This is
called the migration support. Of course, to be able to use the LPM a minimum of
two systems is necessary.

Managed system’s capabilities

The Power Hypervisor is the foundation of the PowerVM. The combination of
Power processor architecture and the Power Hypervisor offers multiple
capabilities among micro-partitioning, virtual processors, virtual Ethernet, virtual
SCSI, virtual console. It is, in essence, always active and cannot be de-activated.
For more information about the Power Hypervisor, refer to PowerVM
Virtualization on IBM System p: Introduction and Configuration Fourth Edition,
SG24-7940

To perform LPM migrations, the servers must be enabled to do so. Therefore, a

valid Advanced Power Virtualization Enterprise Edition or APV EE licence needs
to be activated on both the source and target servers. Those capabilities must at
least be active for:
› VIOS Capable TRUE
› Active Partition Mobility TRUE
› Inactive Partition Mobility TRUE
276 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 7-1 shows one possible output.

Figure 7-1 Server capabilities

Notes:
› If the capability you are looking for is not displayed, this means that your
HMC is down level or the HMC does not have the appropriate fix pack. See
7.4.2, “HMC requirements” on page 285.
› If the capability you are looking for is shown as False, the Power
Hypervisor enablement has not been performed or the firmware is
downlevel.

If your server is not configured with the Enterprise Edition, you need to enable
your system. Contact your IBM representative to do so.
Chapter 7. Live Partition Mobility 277
Processor clock speed
LPM can be used between managed systems with differing clock speeds. For
example, the source server can have 4.7 GHz processors and the target can be
configured with 5.0 GHz processors. This is also known as processor
compatibility mode.

A processor compatibility mode is a value assigned to a logical partition by the

Power Hypervisor, which specifies the processor environment on which the
logical partition can successfully operate. Processor compatibility mode can be
set into the partition’s profile. You can configure modes as shown in Figure 7-2 on
page 279, as follows:
› POWER6 mode or default mode
› POWER6 enhanced mode
This mode provides additional floating point instructions to applications
running on the logical partition.
› POWER5 mode
Typically when operating system’s version installed on the partition does not
support POWER6 processor mode.
278 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 7-2 Processor compatibility mode

The processor compatibility mode is checked by the managed system across the
partition’s profile when the partition is activated and determines whether the
installed operating system supports this mode. If not, the partition uses the most
fully featured mode that is supported by the operating system.

Note: Processor compatibility mode cannot be changed dynamically. You

need to shut down the partition, modify the partition’s profile to the desired
mode, and restart the partition.
Chapter 7. Live Partition Mobility 279
 Note: A POWER6 processor cannot emulate all features of a Power5
processor.

The compatibility mode is important when using LPM. When you move a logical
partition from one system to another that has a different processor type, the
processor compatibility mode enables that logical partition to run in a processor
environment on the destination system in which it can successfully operate.

For more information, see PowerVM Virtualization on IBM System p: Introduction

and Configuration Fourth Edition, SG24-7940.

Time-of-Day
Although not mandatory, it is suggested to synchronize the time-of-day between
the source and the destination system, be it for active or inactive partition
mobility. Therefore, a new attribute, the Time Reference, is available in the
settings tab of the properties of the partition. Any VIOS partition can be
designated as a Time Reference Partition (TRP). Values for the TRP is enable or
disable (default). Changes take effect as soon as you click OK.

Important: It is important to have a the complete environment’s time and date

synchronized for DB2 to remain coherent. You can imagine the problems that
can arise when, after the migration of the logical partition from one system to
another, the time and date suddenly went back in the past. This causes a
major database corruption.

Notes:
› Time synchronization is a suggested step in the active partition migration.
If you chose not to synchronize using the TRP attribute, the source and
target systems synchronizes the clocks during the migration process from
the source server to the destination server.
› The TRP capability is only supported on systems capable of active
partition migration.
› Other TRPs are supported by server. The longest running TRP is
recognized as the TRP of the system.
› Ensure that the HMC and all the partitions have the same date and time to
avoid discordances and errors when migrating to another system.
280 Best Practices for DB2 on AIX 6.1 for POWER Systems
Managed system’s firmware level
Firmware level of both systems acting in a LPM needs to be at a minimum level
of EH330_046 for Power System p595, EM320_31 for Power Systems p570
servers or EL320_040 for Power Systems p520 and p550 servers. PowerBlades
systems need a firmware level of Es320 or later.

Recommendation: It is suggested to have your system at the last firmware

level, because it contains the last features and options that you can use and
the latest correctives.

Refer to the Power code matrix at the following Web page for more
information:

http://www14.software.ibm.com/webapp/set2/sas/f/power5cm/power6.html

Note: Firmware level of source and target systems might defer. However, the
level of the source system firmware must be compatible with the target server.

Refer to the following Web page for more information:

http://www14.software.ibm.com/webapp/set2/sas/f/pm/migrate.html

To upgrade the firmware of your managed system, refer to the IBM fixes web site:
http://publib.boulder.ibm.com/infocenter/systems/scope/hw/index.jsp?top
ic=/ipha5/fix_serv_firm_kick.htm

Recommendation: Firmware level E*340_039 is recommended for all

systems.

Available memory on destination system

Ensure that the destination system has as at least as much memory as the
source system to support the migration. If necessary, use the DLPAR capability
of your system to free memory.

Available processor on destination system

Ensure that the destination system has at least as many processors or processor
unit resources to support the migration. Because the DB2 client profile from the
source system is copied to the destination system, ensure that you have the
requested dedicated processors or processing units to re-construct the DB2
client profile on the destination server.
Chapter 7. Live Partition Mobility 281
Existing profile
Determine the name of the logical partition profile for the mobile partition on the
destination system. This is an optional step. As part of the migration process, the
HMC creates a new migration profile containing the partition’s current state. You
must also verify that no Logical Host Ethernet Adapter (LHEA) devices are
configured, because these are also considered as physical I/O. Inactive migration
is still possible if LHEA are configured. 78 IBM PowerVM LPM replaces the
existing profile that was last used to activate the partition.

Also, if you specify an existing profile name, the HMC replaces that profile with
the new migration profile. If you do not want the migration profile to replace any of
the partition’s existing profiles, you must specify a unique profile name. The new
profile contains the partition’s current configuration and any changes that are
made during the migration.

When migrating a mobile partition to another system, an LPAR with the same
name as the source LPAR might not exist on the destination system. Check the
HMC console or through the command line on the HMC using the lssyscfg
command. See Example 7-1 as an example.

Example 7-1 Output of the lssyscfg command

hscroot@:~> lssyscfg -r lpar -F name,msp,state -m

PRD913B-8204-E8A-SN060E9B2
VIOS_Beta,0,Not Activated
VIOS_Alpha,0,Not Activated
LPAR13-Etl,0,Not Activated
LPAR12_Data,0,Running
LPAR11_Data,0,Running

hscroot@:~> lssyscfg -r lpar -F name,msp,state -m

PRD913B-8204-E8A-SN0653E02
lpar1-data,0,Running

In Example 7-1, each defined LPAR is only defined on one single system.
282 Best Practices for DB2 on AIX 6.1 for POWER Systems
Managed systems’ connectivity
Next, we discuss managed systems’ connectivity.

Single or dual HMC setup

When going for LPM, both the source and the target servers need to be
connected to the same HMC or redundant pair of HMCs, as illustrated in
Figure 7-3. They also need to be connected on the same network so they can
communicate with each other. Only HMC-to-HMC managed systems or
IVM-to-IVM managed systems can intervene in the migration.

Figure 7-3 Dual HMC setup

Use the lssyscfg command to verify that both servers are seen by the same
HMC, as shown in Example 7-2.

Example 7-2 Output of the command lssyscfg

hscroot@:~>lssyscfg -r sys -F name,type_model

PRD913B-8204-E8A-SN060E9B2,8204-E8A
PRD913B-8204-E8A-SN0653E02,8204-E8A

Separate HMC
Using one HMC per server is another supported environment for LPM. This is
known as Remote LPM. Pre-requisites for Remote LPM are covered in 7.4.2,
“HMC requirements” on page 285.

Note: For information about remote partition mobility, refer to IBM Redbooks
publication IBM PowerVM Live Partition Mobility, SG24-7460.
Chapter 7. Live Partition Mobility 283
Resource monitoring and control (RMC) is a feature that can be configured to
monitor resources such as disk space, CPU usage, and processor status and
allows performing an action in response to a defined condition. Remote migration
operations require that each HMC has RMC connections (see “RMC
connections” on page 298) to its individual system’s VIOSs and a connection to
its system’s service processors. The HMC does not have to be connected to the
remote system’s RMC connections to its VIOSs, nor does it have to connect to
the remote system’s service processor

Logical Memory Block (LMB)

The logical memory block, LMB, must be identical on both the source and the
target server. The default LMB size depends on the amount of memory installed
in the CEC. It varies between 16 MB and 256 MB.

Use the following guidelines when selecting logical block sizes:

› On systems with a small amount of memory installed (2 GB or less), a large
logical memory block size results in the firmware consuming an excessive
amount of memory. Firmware must consume at least 1 logical memory block.
As a general rule, select the logical memory block size to be no greater than
1/8th the size of the system’s physical memory.
› On systems with a large amount of memory installed, small logical memory
block sizes result in a large number of logical memory blocks. Because each
logical memory block must be managed during boot, a large number of logical
memory blocks can cause boot performance problems.
As a general rule, limit the number of logical memory blocks to 8 K or less.

Note: The logical memory blocksize can be changed at run time, but the
change does not take effect until the system is restarted.

Battery power
Ensure that the target system is not on battery power. If so, bring the target
system in a stable state prior to migrate.

Note: The source system can run on battery power even if you want to migrate
your partition to another system.
284 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.4.2 HMC requirements
Table 7-3 summarizes HMC requirements.

Table 7-3 HMC requirements

HMC tasks Actions to undertake and remarks For details see

HMC model › 7310-CR2 desktop “HMC model and version”

› 7310-C03 rack-mount on page 285

HMC level (single or redundant › 7.3.2.0 or later “HMC model and version”
dual environment) › fix MH01062 on page 285

HMC level in remote LPM 7.3.4 or later “HMC model and version”
on page 285

Non-symmetric environments › 7.3.5 or later “HMC model and version”

› VIOS 2.1.2.10 FP22 on page 285

HMC connectivity Interconnected “HMC model and version”

on page 285

HMC model and version

Two types of HMC can be used in an LPM environment. You can use either a
desktop model 7310-CR2 or later, or a rack-mounted model 7310-C03. See
Figure 7-4.

Figure 7-4 Desktop and rack-mount HMCs

Both servers need to be connected to the same HMC or redundant pair of HMCs
as illustrated in Figure 7-3 on page 283. They also must be connected on the
same network so they can communicate with each others, as shown in
Figure 7-3 on page 283.
Chapter 7. Live Partition Mobility 285
 Notes:
› HMC can perform multiple LPM migrations simultaneously.
› IVM is also supported to perform LPM migrations.

For the LPM to be successful in a single or dual redundant HMC environment,

the HMCs’ version need to be at a minimum level of 7.3.2 or later with required fix
MH01062. Each HMC can see all the managed systems. Both HMC in a
redundant dual HMC environment need to be at the same level.

For information about how to upgrade your HMC, refer to the following Web page:
http://www-933.ibm.com/support/fixcentral/

Tip: Migrations across a non-symmetric HMC configuration is possible if you

meet the following requisites:
› HMC is at level 7.3.5 or later
› GUI “override errors” option or command line with -force flag

For example, migration of a logical partition to a destination server whose

VIOS does not provide the same level of redundancy as the source system.
› Source system has two redundant VIOS serving the mobile partition
through two SCSI channel
› Destination system only has one VIOS with only one HBA.

In other configurations, the source server can be connected to one HMC,

whereas the target server can be connected to another HMC. LPM operations
are possible. This situation where a logical partition needs to migrate from one
server to another server is known as remote LPM. To perform a remote LPM
ensure that:
› The source and target HMCs are connected to the same network so they can
communicate with each other.
› The source and target HMC's version need to be at level 7.3.4 or later.
› To display the HMC version, use the lshmc -V command,
› A secure shell (SSH) must be established between both the source and the
target HMC. For more information about how to set up SSH keys
authentication refer to the following Web page:
http://publib.boulder.ibm.com/infocenter/powersys/v3r1m5/index.jsp
› The migration of the mobile partition is managed by the HMC containing the
mobile partition.
286 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.4.3 IVM requirements
IVM is closely tied to the VIOS media support you have. Each VIOS media
support contains indeed the IVM package you need to install on your system to
manage your partitions.

Therefore the requirements on IVM are tied to the VIOS level requirements. You
can refer to VIOS Requirements, part “VIOS profile” on page 288 for more
information.

Note: IVM actually helps to mask the complexity of the underlaying

commands that run at VIOS level. However for particular reason among
requested by the IBM support, you might need to run commands at VIOS
level.

7.4.4 VIOS requirements

Table 7-4 lists the VIOS requirements.

Table 7-4 VIOS requirements

VIOS asks Actions to undertake and remarks For details, see

VIOS level › VIOS 1.5.2.1-FP11 “Operating system

› Fixes level” on page 288
VIOS profile › Mover Service Partition (not needed for inactive migration) “VIOS profile” on
› Destination server page 288
– Not on power battery
– Not containing existing profile
– Has sufficient resources (CPU, memory, # virtual slots)
› CPU & memory
› VASI Interface (not needed for inactive migration)
› Virtual SCSI
› IVE
› Virtual Ethernet
› Virtual slots ID higher than 11
› Time and date synchronization (optional for active
migration, not needed for inactive migration)

Dual VIOS › Symmetric environment recommended “Dual VIOS” on

› Support for non-symmetric environment if VIOS page 292
2.1.2.10-FP22

Disk mapping Only disk mapping “Disk mapping to

to client client partition” on
page 293
Chapter 7. Live Partition Mobility 287
For VIOS settings recommendations, refer to Chapter 6, “Virtual I/O” on
page 241.

Operating system level

The VIOS needs to be at a minimum level of 1.5 with its limited released fixes
(VIOS 1.5.2.1-FP11) to migrate a logical partition successfully.

Tip: Version 2.1.2.10 FP22 of the VIOS introduced a new functionality that
consists of the preservation of the customized VTD names, support for
non-symmetric environments, and multiple IP addresses for the VIOS
configured as a MSP partition.

All VIOS need to be at the same release level.

For more information about how to change your current VIOS level, refer to the
following Web page:
http://www14.software.ibm.com/webapp/set2/sas/f/vios/download/home.html

Tip: The client profile must only be created on the source system. The profile
on the destination system is created automatically during the migration
process.

VIOS profile
This section discusses the VIOS profile.

Mover Service Partition

Mover Service Partition (MSP) is an attribute of the VIOS partition. It enables the
VIOS partition to allow the function that asynchronously extracts, transports, and
installs a partition state. Two mover service partitions are involved in an active
migration: one on the source system and one on the destination

There must be at least one VIOS declared as a mover service partition per
system. A mover VIOS partition is the specific ability of that partition to interact
with the Power Hypervisor and to migrate a mobile partition from the source
VIOS to the target VIOS.

The mover service partition attribute can be configured either when you
configure your VIOS profile (refer to IBM Redbooks publication PowerVM
Virtualization on IBM System p: Introduction and Configuration Fourth Edition,
SG24-7940 for more information) or after the VIOS has been configured by
changing its properties. You can proceed as described in Figure 7-5 on page 289
by editing the properties of the VIOS partition and selecting the Mover Service
Partition box.
288 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Note: Mover service partitions are not used during inactive migration partition.

Figure 7-5 VIOS Mover Service Partition capability for LPM

Note: If the mover service partition is disabled on either the source or

destination VIOS, the mobile partition can participate only in inactive partition
mobility.

Tip: If you have multiple network interfaces configured on an MSP, you can
chose, through the HMC command line, which IP address the mover uses to
transport the mobile partition’s data. For this to work, you need VIOS level
2.1.2.10-FP22 or later and HMC level 7.3.5 or later.
Chapter 7. Live Partition Mobility 289
CPU and memory requirements
VIOS requires modest additional CPU and memory. Assuming that CPU cycles
required to host VIOS are available, there is modest performance impact with
VIOS from the client partition perspective. Check the IO latency impact to your
application from a host perspective (DB2 server).

During the migration of a LPAR, each VIOS of both the source and the
destination system need extra CPU to manage the migration. Uncapping the
VIOSs assures the necessary additional CPU to be given to each of them.

The following list details considerations regarding VIOS CPU and memory
requirements:
› Allocating dedicated CPU to a VIOS partition results in faster scheduling of
I/O operations and improved Virtual I/O performance.
› When the overall pool use is high, and using uncapped shared micropartition
for VIOS, allocating less than 1.0 Entitled CPU (ECPU) to VIOS partition
might result in longer latency for IO.
› Memory required for a VIOS partition is insignificant. Allocating more memory
does not result in any performance improvement.

Tip: To size the VIOS CPU appropriately, take the following factors into
consideration:
› LPM requires additional CPU during LPAR mobility.
› Shared uncapped CPU gives you the flexibility to acquire additional CPU
units whenever needed.

For more information about VIOS CPU and memory requirements, refer to 6.4,
“VIOS sizing” on page 257.

Destination server requisites

A destination system might not contain the existing LPAR name of the LPAR on
the source system.

The destination system needs to have sufficient resources to host the migrating
partition, not only those such as CPU processing units, and memory, but also
virtual slots. Virtual slots are needed to create the required virtual SCSI adapter
after the mobile partition has moved to the destination system. You might check
for the maximum number of virtual adapters and the actual number of configured
virtual adapters.
290 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Recommendations:

- It is suggested to have symmetric configuration between source and

destination systems. (see the Tip in “Dual VIOS” on page 292).

- It is suggested that the destination system not be running on a battery. If it

does, the situation needs to be corrected before the start of the migration.

VASI interface
The Virtual Asynchronous Service Interface (VASI) provides communication
between both mover service partitions and the Power Hypervisor to gain access
to the partition state. This VASI interface is automatically created when the VIOS
is installed and the VIOS is declared as a mover service partition.

Tips:
› To list partition migration information, use the lslparmigr command.
› VASI interface must be available and is mandatory for active LPM. It is not
required for inactive partition mobility.

Virtual SCSI interface

Virtual SCSI interfaces need to be defined as Desired (not required) in the VIOS
profile to allow the migration process to remove those interfaces during a
migration.

When configuring the virtual SCSI interface, it is suggested to create a virtual link
between the VIOS and the logical partition. This can be done by selecting Only
selected client partition can connect in the VIOS profile. Moreover, the VIOSs
on both the source and the destination systems must be capable of providing
virtual access to all storage devices that the mobile partition is using.

Integrated virtual Ethernet IVE

If you plan to use IVE on your VIOS, ensure it has been configured in
promiscuous mode. This mode guarantees that only your particular partition can
use the complete port of the associated port group of your IVE card.

For more information about configuring IVE, refer to Integrated Virtual Ethernet
Adapter Technical Overview and Introduction, REDP-4340.

Virtual Ethernet
Virtual Ethernet adapters are created in the VIOS’ profile to ensure the
communication between the VIOSs and the logical partitions configured on a
same system.
Chapter 7. Live Partition Mobility 291
It is mandatory to create shared Ethernet adapters (SEA) on each the source
and the destination server to bridge to the same Ethernet network used by the
mobile partition. For more information about SEA settings, refer to “SEA
configuration” on page 303.

Virtual slots
Make sure sufficient virtual slots are available on the target system.

Note: Any user defined virtual slot must have an ID higher than 11.

Time and date synchronization

Time and date between VIOSs need to be synchronized. An attribute, the Time
Reference, is available in the settings tab of the properties of the partition. Any
VIOS partition can be designated as a Time Reference Partition (TRP). Values
for the Time Reference is enable or disable (default). Changes take effect as
soon as you click OK.

Time-of-day synchronization is optional for both active and inactive partition

migration, but it is a suggested step in the active partition migration. If you chose
not to synchronize using the TRP attribute, the source and target systems
synchronizes the clocks during the migration process from the source server to
the destination server.

Notes:
› The TRP capability is only supported on systems that are capable of active
partition migration.
› Other TRP are supported by servers. The longest running TRP is
recognized as the TRP of the system.
› Ensure that the HMC and all the partitions have the same date and time to
avoid discordances when migration to another server.

Dual VIOS
For more information about dual VIOS setups, refer to 6.3.1, “Dual VIOS” on
page 251.

Tip: Non-symmetric configurations are possible when using partition mobility,

even though this is not a recommended situation.

If you are in such a situation, only VIOS version 2.1.2.10-FP22 or later is

allowed as operating system and HMCs need to be at level 7.3.5.0 or later for
LPM to function properly.
292 Best Practices for DB2 on AIX 6.1 for POWER Systems
Disk mapping to client partition
Any client partition willing to act in a live partition needs to have its mapped disk
of a whole LUN. Logical volumes mapping is not supported with LPM.

In other words, the disks zoned on a VIOS (these are LUNs on the storage box
and seen as hdisk on the VIOS) need to be mapped entirely to the logical
partition that needs to be migrated from one source system to another.

Moreover, these LUNs must be accessible to VIOSs on both the source and the
destination server. Other parameters at disk level and HBA level need to be
adapted. For more information about those parameters, refer to 3.4, “Tuning
storage on AIX” on page 115.

Tips:
› For all disks in a mobile logical partition, ensure the following settings for
the HBAs on all of your VIOSs:
– reserve_policy: no_reserve policy
– hcheck_interval: 20
– fc_err_recov: fast_fail
– dyntrk: yes
› For all HBAs in VIOSs ensure the following settings:
– fc_err_recov: fast_fail
– dyntrk: yes.
Chapter 7. Live Partition Mobility 293
7.4.5 LPAR requirements
Table 7-5 lists the LPAR requirements for active migration.

Table 7-5 LPAR requirements for active migration

LPAR tasks Actions to undertake and For details see
remarks

LPAR profile Unique name “LPAR name” on page 296

Operating system › AIX 5300-07-01 “Operating system requirements”

› AIX 6100-00-01 on page 296
› RedHat RHEL1a or later +
kernel security update
› Slesb 10 or later + kernel
security update

Processor setting › Dedicated / Shared “Processor settings” on page 298

› Capped
› Uncapped

RMC connections For active migrations: active “RMC connections” on page 298
RMC connection, rsct daemons
running

Physical I/O Only virtual adapters “Physical I/O” on page 299

MAC address Unique across both systems “Network” on page 300

IVE (HEA) adapter No IVE / HEA adapters “Integrated Virtual Ethernet (IVE)”
on page 300p

Partition workload group Might not be part of partition “Partition workload group” on
workload group page 300

Virtual serial adapter Only serial IDs 0 & 1 “Virtual Serial adapter” on page 301

Barrier Synchronization Number of arrays = 0 “Barrier Synchronization Register

Register (BSR) (BSR)” on page 301

Huge pages Requested huge pages = 0 “Huge pages” on page 301

Redundant Error Path Disabled “Redundant Error Path Reporting”

on page 303
a. RHEL: RedHat Enterprise Linux
b. SLES: Suze Linux Enterprise Server
294 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Table 7-6 lists the LPAR requirements for inactive migration.

Table 7-6 LPAR requirements for Inactive migration

LPAR tasks Actions to undertake and remarks For details see

LPAR profile Unique name “LPAR name” on page 296

Operating system › Prior AIX 5L if OS supports “Operating system

POWER6 and virtual devices requirements” on page 296
› 5200-10 or later (only for 9117
MMA systems)
› 5300-06 or later
› 6100 or later
› RedHat RHEL1 or later + kernel
security update
› Sles 10 or later + kernel security
update

RMC connections Not needed “RMC connections” on

page 298

Physical I/O Can have dedicated I/O “Physical I/O” on page 299

MAC address Unique across both systems “Network” on page 300

IVE (HEA) adapter Can be used “Integrated Virtual Ethernet

(IVE)” on page 300

Partition workload group Might be part of partition workload “Partition workload group” on
group page 300

Virtual serial adapter Only serial ID’s 0 & 1 “Virtual Serial adapter” on
page 301

Barrier Synchronization Can be used “Barrier Synchronization

Register (BSR) Register (BSR)” on page 301

Huge pages Can be used “Huge pages” on page 301

Redundant Error Path Can be used “Redundant Error Path

Reporting” on page 303
Chapter 7. Live Partition Mobility 295
LPAR name
Partition mobility is only possible for those partition profiles configured to run AIX
or Linux operating systems. Even though VIOSs are configured as a Mover
Service Partition (MSP), they cannot be migrated.

Note: The target server might not contain a LPAR with the same name as the
partition you want to move.

For the migration to be successful, the LPAR name from the migrating partition
must not exist on the destination system. You can, however, determine a new
name for your partition. The HMC creates a new profile containing the partition’s
current state, the configuration, and any changes that are made during the
migration.

Note: For a logical partition to participate in an active partition migration, it

cannot have any physical I/O. All I/O must be virtual. If the mobile partition has
required or physical I/O, it can participate in inactive partition migration.

Tip: Devices configures as Desired can be dynamically removed from a

partition’s profile. Others marked as Required require you to change the profile
and shutdown. Reboot your partition for the changes to take effect.

Operating system requirements

Operating system requirements for logical partitions differ depending on the type
of migration you want to perform: active or Inactive migration (see Table 7-7 on
page 297 and Table 7-8 on page 297).

Notes:
› Active partition migration
This is the ability to move a running logical partition with its operating
system and application from one system to another without interrupting the
service / operation of that logical partition.
› Inactive partition migration
This is the ability to move a powered off logical partition with its operating
system and application from one system to another.
296 Best Practices for DB2 on AIX 6.1 for POWER Systems
Table 7-7 lists the operating system level needed for active partition mobility.

Table 7-7 Operating system level needed for active partition mobility
OS level Active migration

AIX 5L Version 5.3 Technology Level 7 or later

AIX 6 Version 6.1 or later

RedHat Enterprise Version 5 (RHEL5) Update 1 or later

Edition Kernel security update

RedHat Enterprise Version 10 (SLES 10) Service Pack 1 or later

Edition Kernel security update

Table 7-8 lists the operating system level needed for inactive partition mobility.

Table 7-8 Operating system level needed for inactive partition mobility
OS level Inactive migration

Prior AIX 5L Any version, as long as supported on POWER6, supports

virtual devices

AIX 5L Version 5.2 Technology Level 10 or later (Only on 9117 MMA

Systems)
Version 5.3 Technology Level 6 or later

AIX 6 Version 6.1 or later

RedHat Enterprise Version 5 (RHEL5) Update 1

Edition Kernel security update

Suse Enterprise Version 10 (SLES 10) Service Pack 1

Edition Kernel security update
Chapter 7. Live Partition Mobility 297
 Notes:
› For earlier versions of the operating systems, inactive partition mobility is
still possible. Your operating system, AIX or Linux, needs to support virtual
devices.
› Inactive partition migration is less restrictive in terms of pre-requisites than
the active partition migration. The following list outlines the elements that
you can configure for inactive partition mobility:
– Dedicated I/O (I/O is removed from the partition before the migration
occurs.)
– Barrier Synchronization Register (BSR)
– Huge pages
– Some dynamic changes (for example, partition workload groups, MSP,
time-reference and VASI)
– Static changes (BSR and redundant error path reporting)
– No need to have a MSP, neither on source nor on target system
– No need to have a VASI adapter on the VIOSs
– No need to synchronize the time-of-day clocks
– No need to have RMC connection

Processor settings
Configuring processors is of high importance, be it on VIOSs or logical partitions.

For optimal balanced performance with an uncapped shared processor LPAR,

we suggest setting a maximum two virtual processors per physical processor: 1
physical CPU ==> max 2VP.

For capped shared processor LPAR, set the number of virtual processor equal to
a round-up of the entitled capacity.

Note: For a more detailed information about processor settings, refer to

Chapter 5, “LPAR considerations” on page 225.

RMC connections
Resource monitoring and control (RMC) is a feature that can be configured to
monitor resources such as disk space, CPU usage, and processor status. It
allows performing an action in response to a defined condition. It is actually
technically a subset function of the Reliable Scalable Cluster Technology
(RSCT). For more information about RMC, refer to IBM Redbooks publication A
Practical Guide for Resource Monitoring and Control (RMC), SG24-6615.
298 Best Practices for DB2 on AIX 6.1 for POWER Systems
Prior to an active migration of your logical partition, ensure that RMC connections
are established between:
› VIOSs of both systems intervening in the mobility.
› Mobile partition and the both the source and destination systems.

Tip: Use the lsrsrc IBM.ManagementServer command to verify the RMC

connections are active and check for ManagerType=HMC. Alternatively, you
can use the lspartition -all command on the HMC and check against your
partition for Active:<3>.

Tips:
› To re-synchronize the RMC connection run the
/usr/sbin/rsct/bin/refrsrc IBM.ManagementServer command. Running
this command suppresses the wait time after synchronization.
› Your LPAR is not listed with the lspartition command if it is not in running
state (it is not powered on).
› RMC needs about five minutes to synchronize after network changes or
partition activation.

Physical I/O
When going for LPM, no physical or required adapters can be configured in the
mobile partition’s profile. All I/O must be virtual. However, if the mobile partition
has physical or dedicated adapters, it can participate in an inactive partition
migration. Physical adapters marked as Desired can be removed dynamically
with a dynamic LPAR operation. For more information about using DLPAR, refer
to PowerVM Virtualization on IBM System p: Introduction and Configuration
Fourth Edition, SG24-7940.

Tips:
› You can chose to use physical adapter in your mobile partition. If you do so
and want an active partition mobility, move your physical adapters to virtual
adapters prior the migration can occur. Downtime might be necessary to
perform these tasks.
› For your particular LPAR running your DB2 database, it is suggested not to
share the bandwidth of your I/O card with other partitions to avoid
unnecessary overhead on the VIOS. Evaluate the need to share it, keeping
in mind performance.
Chapter 7. Live Partition Mobility 299
Network
The logical partition’s MAC address need to be unique across both systems. You
might consider using the netstat or entstat command to check this.

Moreover, the mobile partition’s network must be virtualized using one or more
VIOSs. Logical Host Ethernet Adapters (LHEA) are not supported for a partition
to be active in a migration process. see “Integrated Virtual Ethernet (IVE)” on
page 300 for more information about Integrated Virtual Ethernet.

Notes:
› Systems configured with IVM have only one VIOS.
› For systems configured with HMC, it is suggested to configure two VIOSs.

Integrated Virtual Ethernet (IVE)

An integrated virtual Ethernet adapter (IVE), also known as a Host Ethernet
adapter, is available in a two or four port integrated Ethernet card and is directly
attached to the Power Hypervisor. The power Hypervisor can create up to 32
logical Ethernet ports that can be given to the logical partition. These ports are
seen as LHEA and provide the logical partitions with a virtual Ethernet
communication link without the necessity to use a VIOS to virtualize an Ethernet
card given to it.

If you plan to use the LPM capabilities of your Power System, you cannot use a
logical host, as LHEA adapters are considered as physical adapters.

Partition workload group

For a mobile partition to migrate from one source system to a destination system,
the mobile partition must not be part of a partition workload groups.

Notes:
› A partition workload group is a group of logical partitions whose resources
are managed collectively by a workload management application. A
partition workload group identifies a set of partitions that reside on the
same system.
› Workload management applications can balance memory and processor
resources within groups of logical partitions without intervention from the
HMC (Hardware Management Console). Workload management
applications are installed separately and can be obtained from a solution
provider company.
300 Best Practices for DB2 on AIX 6.1 for POWER Systems
For more information about partition workload group and workload management
applications, refer to Chapter 5, “LPAR considerations” on page 225.
Virtual Serial adapter
Virtual serial adapters are often used for virtual terminal connections to the
operating system. The two first serial adapters (adapter ID’s 0 and 1) are
reserved for the HMC.

Validate that no physical adapters are in the mobile partition and that no virtual
serial adapters are in virtual slots higher than 1. In other words, the only
exception for virtual serial adapter is for the virtual terminal connection.
Barrier Synchronization Register (BSR)
Barrier synchronization registers provide a fast, lightweight barrier
synchronization between CPUs. This facility is intended for use by application
programs that are structured in a single instruction, multiple data (SIMD) manner.
Such programs often proceed in phases where all tasks synchronize processing
at the end of each phase. The BSR is designed to accomplish this efficiently.
Barrier synchronization registers cannot be migrated or re-configured
dynamically. Barrier synchronization registers cannot be used with migrating
partitions.

Notes:
› BSR can be used in inactive partition migration.
› Disabling BSR cannot been changed dynamically. Modify the partition’s
profile and shutdown (not reboot) the partition for changes to take effect.
› If your migrating partition contains dedicated physical resources, it is
mandatory to move those physical resources to virtual resources prior to
migrating or the migration fails.

Huge pages
Huge pages can improve performance in specific environments that require a
high degree of parallelism. The minimum, desired, and maximum number of
huge pages can be specified to assign to a partition when you create a partition
profile.

However, for a logical partition to participate in active partition migration, it cannot

use huge pages. For more information about Huge pages, refer to 2.1.4, “Large
page considerations” on page 46.

Note: If your mobile partition does use huge pages, it can participate in an
inactive partition migration.
Chapter 7. Live Partition Mobility 301
Huge page settings can be checked on the HCM console as follows:
1. Select your system in the navigation area.
2. Choose Properties.
3. Select the Advanced tab
4. Verify that the Current Requested Huge Page Memory field indicates 0 (zero).
See Figure 7-6.
If this field is not equal to zero, only an inactive partition mobility works.

Figure 7-6 Huge page settings

Note: If you need to use huge pages and active partition mobility, you need to
set to 0 all fields relative to huge page memory of your mobile partition for the
migration to be successful. As this is not a dynamic operation, you need to
modify the partition’s profile and shutdown (not reboot) the partition for the
changes to take effect.
302 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Redundant Error Path Reporting
This indicates whether the logical partition is set to report server common
hardware errors to the HMC. The service processor is the primary path for
reporting server common hardware errors to the HMC. Selecting this option
allows you to set up redundant error reporting paths in addition to the error
reporting path provided by the service processor. You can change this setting by
activating the logical partition using a partition profile set to enable redundant
error path reporting. For mobile partitions, this profile attribute must be
de-activated.

7.4.6 Network requirements

Table 7-9 shows the network requirements.

Table 7-9 Network requirements

Tasks Actions to undertake For details see
Remarks

SEA configuration › For active partition mobility: “SEA

SEA failover mechanism configuration” on
› For inactive partition page 303
mobility: SEA

Network access Communication between both “Access to the

systems necessary network” on
page 305

IVE/LHEA adapter › For active partition “LHEA adapter”

migration: not supported on page 305
› For inactive partition
mobility: supported

The network is important to consider in an environment where you want to use

the migration capabilities of your servers.

SEA configuration
You need to complete several tasks to ensure your network is ready for the
migration.

You have to create a shared Ethernet adapter (SEA) on both of your source and
destination system VIOSs. The SEA bridges your external network to your
internal virtual network (Layer-2 bridge). It provides the ability to your client
partitions to share one physical Ethernet adapter. When using dual VIOS setup
in an HMC environment, the network can be made highly available by creating
the SEA failover mechanism, as shown in Figure 7-7 on page 304.
Chapter 7. Live Partition Mobility 303
 Notes:
› SEA can only be hosted on VIOSs.
› VIOS running on IVM cannot implement the SEA failover mechanism
because it can only contain one single VIOS. For more information about
SEA failover mechanism, refer to IBM Redpaper IBM System p Advanced
POWER Virtualization (PowerVM) Best Practices, REDP-4194.

SEA can be built of physical interfaces or IVE cards. If you want to use the IVE
card, ensure that the IVE port you are using is set to promiscuous mode. This
mode ensures that the complete port is dedicated to your VIOS and that no other
partition is able to use it. For more information about IVE, refer to Integrated
Virtual Ethernet Adapter Technical Overview and Introduction, REDP-4340.

Figure 7-7 Sample SEA failover setup with Etherchannel and dual VIOS setup
304 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Recommendation: It is suggested to configure your IP address of your VIOS
on an additional virtual adapter that you need to create, apart from the SEA
failover mechanism.

Access to the network

The source and target systems need to communicate with each other. You must
give access to both systems to the same network segment and VLAN. Moreover,
partitions and HMC also need to exist on the same network.

LHEA adapter
When configuring your mobile logical partition, make sure that no LHEA device is
configured. LHEA devices are considered as physical devices. Active partition
mobility requires no physical adapter to migrate successfully. However, inactive
partition mobility can contain LHEA devices if you want to use these.

For more information about Integrated Virtual Ethernet adapters (IVE / HEA)
refer to WebSphere Application Server V6.1: JMS Problem Determination,
REDP-4330.
Chapter 7. Live Partition Mobility 305
7.4.7 Storage requirements
Table 7-10 lists the storage requirements.

Table 7-10 Storage requirements

Tasks Actions to undertake For details
Remarks see

LUN creation on SAN storage › Create LUN in particular “LUN creation

box RAID level on SAN
› Divide LUN into desired storage box”
capacity on page 310
› Map to host group

Disks and HBA parameters Disks parameters: “Disk and HBA

› algorithm=round_robin parameters”
› hcheck_interval=20 on page 310
› max_transfer=
› reserve_policy=no_reserve
› queue_depth=(depends on
the storage vendor and type)
HBA parameters:
› dyntrk=yes
› fc_err_recov=fast_fail
› max_xfer_size=
› lg_term_dma=
› num_cmd_elems=

Disk mapping versus LUN Only disk mapping supported “Disk mapping
masking versus LUN
masking on
VIOS” on
page 314

Note: Some of the recommended parameters depend upon the storage

subsystem you are using. For more information about those specific
parameters, refer to 3.4, “Tuning storage on AIX” on page 115.

LPM functions under a specific set of rules. As we have already discussed, it is

licensed under PowerVM APV Enterprise Edition. All storage and networking
resources must be virtual, as no physical device can be moved. So, any
supported virtual storage device (SAN, iSCSI, and so forth) can be configured to
provide storage space to your mobile logical partition. The virtual storage device
cannot be a logical volume contained within a volume group assigned to the
VIOS. To enable multiple system access, the reserve policy must be set to allow
306 Best Practices for DB2 on AIX 6.1 for POWER Systems
concurrent access. This is accomplished through the no_reserve policy attribute.
This needs to be performed before the vtscsi device is created, using the
following command:.
$ chdev - dev hdiskn -attr reserve_policy =no_reserve

For our testing we used a DS5300 SAN Storage Solution. We set up transaction
logs and data on separate file systems and, ideally, on external storage systems
on their own dedicated LUNs. In our example, we set up transaction logs under
the /db2_log_bck directory and data under /db2_data. These file systems were
located in the SAN on their own dedicated LUNs. We defined our database paths
as shown using the database creation clause. The database was configured as
an automatic storage database, with the containers placed on /db2_data.
CREATE DATABASE TPCE AUTOMATIC STORAGE YES on /db2_data DBPATH on
/home/db2inst1 USING CODESET ISO8859-1 TERRITORY US COLLATE
USING IDENTITY;

Using automatic storage significantly reduces time spent on maintenance when

there is a need to add storage paths to our database. Automatic storage provides
optimal performance, so it is an ideal choice for storage type. When you need
more storage space or I/O you need to provide additional LUNs from your SAN
system and make them available to your OS. After this, you can add the new
storage path to your database environment. This increase in storage paths for
your database table spaces increases the amount of storage, and I/O capacity.

The following example shows the command for adding two additional storage
paths to our database. Ideally, each storage path is equal in size and lay on its
own file system under dedicated LUNs.
ALTER DB virtdb ADD STORAGE PATH

When the I/O capacity has increased, you might need more CPU capacity to
handle the increased I/O. For an uncapped shared processor logical partition,
this is not a problem as long as there is enough CPU capacity in the shared
processor pool. If the increase in the needed CPU capacity is permanent,
consider increasing the entitled processor capacity for the database logical
partition. With dynamic logical partitioning, this can be achieved without any
effect to service. You must change the value of the desired processor units for the
LPAR. For a Power Systems server environment you must choose between the
virtualized storage and the locally attached storage. Although the virtualized
storage provides easier maintenance, it has a slight overhead. On most systems
this overhead is not noticeable, so we are able to take advantage of the benefits
for easier maintenance when using virtual storage. VIOS makes it easier to add
more storage to your system without a service break. With more than one virtual
server, you are always able to add physical adapters and SAN systems to your
environment without service breaks on your production environment. This ability
Chapter 7. Live Partition Mobility 307
to provide additional storage and I/O resources without service downtime
combined with the ease of use and maintenance of DB2 automatic storage,
makes Power Systems environments ideal for dynamic computing environments
with dynamic resource needs.

For more information about the alternate approaches (NFS, iSCSI and FCP) to
attach storage to a DB2 server, seeIBM DB2 9 on AIX 5L with NFS, iSCSI, and
FCP using IBM System Storage N series, REDP-4250.

A whitepaper covering the testing that was done using DB2 and LPM using a
network attached storage (NAS) over iSCSI can be found at the following Web
page.
https://www-304.ibm.com/partnerworld/wps/servlet/ContentHandler/whitepa
per/power/lpm/use

There is also a demo that has been built from the tests done with DB2 and NAS
over iSCSI. It can be found at the following Web page:
http://www.ibm.com/partnerworld/wps/servlet/ContentHandler/VPAA-7M59ZR

Storage is a crucial point in the setup of an environment that needs to take

benefit of the migration capabilities of Power Systems.
308 Best Practices for DB2 on AIX 6.1 for POWER Systems
 The storage that you define on your SAN storage box must be on an external
SAN storage box, accessible by both the source and the destination systems, as
shown in Figure 7-8.

Figure 7-8 SAN Architecture: Example of Disk Mapping

DB2 performance is heavily dependent on the I/O subsystem performance. To

attain the best possible I/O throughput, the data layouts of database tables
demand special attention from database administrators and system
administrators. The chosen I/O type has great impact on the manageability and
extensibility of the DB2 storage requirements. Therefore, it is critical to consider
workload priorities and to examine trade-offs between disk I/O types. You can
choose between locally attached I/O or VIO, or both, within a partition.

All client logical partition requests are passed down to the VIOS, where it
performs the actual disk I/O operation and returns data directly to a client
partition (no double buffering in the case of Virtual SCSI).
Chapter 7. Live Partition Mobility 309
LUN creation on SAN storage box
LUNs need to be defined, following the best practices for storage configuration,
using RAID techniques as described in 3.3, “Storage hardware” on page 100.

From the array configuration on the SAN, LUNs from equal size are created. A
host group is defined and contains the hosts. Those hosts are referenced by their
world wide port name (WWPN). Host groups contain those hosts that share the
same disks. When migrating a logical partition from a source to a destination
system, all disks defined into that logical partition need to be visible from both the
source and the destination system for the migration to be successful. As for the
network that needs to be accessible by both systems, the SAN disks must be
able to attach to either the source or the destination system. It is one or the other,
not both system at the same time. Therefore, a few parameters need to be set
not only at disk level, but also at Fibre Channel adapter level, as explained in the
next sections.

Disk and HBA parameters

Prior to map disks to logical partition, to change the disk and HBA parameters to
allow both the systems to attach the shared disks.

Here are the suggested settings for the disks parameters (See the list that follows
these settings for explanations about the parameters):
› algorithm: round_robin
› hcheck_interval: 20
› reserve_policy: no_reserve
› max_transfer: see the Note, “Setting the max-transfer size” on the next page.
› queue_depth: This parameter depends on the storage box used and on the
storage box firmware.

More information about these parameters is given in the following list:

› algorithm
The algorithm attribute defines the methodology that the Path Control Module
(PCM) uses to manage I/O across the paths configured for a device.
(failover: all I/O go through one single path and paths are kept in a list to
determine the next path to use, round_robin: all I/O is equally distributed
across enabled paths)
› hcheck_interval
This parameter defines how often the health check is performed on the paths
for a device. The attribute supports a range from 0 to 3600 seconds. When a
value of 0 is selected, health checking is disabled.
310 Best Practices for DB2 on AIX 6.1 for POWER Systems
› reserve_policy
The reserve_policy attribute determines the type of reserve methodology the
device driver implements when the device is opened. It can be used to limit
device access from other adapters, whether on the same system or another
system. (single_path, no_reserve, PR_exclusive, PR_shared)
› max_transfer
The maximum transfer parameter determines the size, in bytes, of the largest
single data transfer request that can be handled by this adapter.

Setting the max-transfer size: If a backing device (such as hdisk or logical

volume) is exported through a vhost adapter, and its max_transfer size is
greater than the max_transfer size of an already associated backing device of
that vhost adapter, the new virtual disk is not presented to the client partition.

You might want to configure the max_transfer size of your newly added disk to
the largest max_transfer size of the existing backing devices. To change the
max_transfer size run the following command:

chdev -dev hdiskx -attr max_transfer=<new size>

In this command, hdiskx represent the disk whose parameters need to be

changed and <new size> is the new size in hex format for your max_transfer
parameter.
› queue_depth
The queue_depth parameter represents the number of requests a disk can
hold in its queue. The default value for IBM disks is three. For non-IBM disks
this value is zero. Changes to this value are immediate and permanent.
Chapter 7. Live Partition Mobility 311
 Setting the queue_depth parameter:

Here are recommendations for setting the queue_depth parameter:

› For storage systems running DS4000 /DS5000 controller firmware
07.10.xx.xx, use the following formula to determine maximum queue depth:
DSxxx queue depth / [(number of hosts) * (number of LUNs per hosts)]

In this formula, DSxxx queue depth is the queue depth of your storage box.
For DS4800 or DS5000, queue depth is 4096.
› Disk queue_depth is aligned on the queue_depth of the storage box.
› The queue depth of the disk on the VIOS must match the queue depth of
the virtual disk on the logical partition.

For more information about setting your storage system, refer to IBM Midrange
System Storage Hardware Guide, SG24-7676 and to 3.3, “Storage hardware” on
page 100.

Fibre Channel devices such as fscsix, need other parameters to be adapted:

The suggested settings for the HBA parameters are as follows (More information
about these parameters can be found in the list after the Note below):
› dyntrk=yes
› fc_err_recov=fast_fail
› max_xfer_size= See Note
› lg_term_dma=See Note
› num_cmd_elems=See Note

Note: As these parameters depend upon the storage type you are using, refer
to Chapter 3, “Storage layout” on page 85 for the optimum values for the
particular storage type you are using.
312 Best Practices for DB2 on AIX 6.1 for POWER Systems
More information about these parameters is given in the following list:
› dyntrk
This parameter enables dynamic tracking and enables the FC adapter driver
to detect when the Fibre Channel N_Port IDD of a device changes. The FC
adapter reroutes traffic destined for that device to the new address while the
device are still online.
› fc_err_recov
AIX supports Fast I/O Failure for Fibre Channel devices after link events in a
switched environment. If the Fibre Channel adapter driver detects a link
event, such as a lost link between a storage device and a switch, the Fibre
Channel adapter driver waits a short period of time, approximately 15
seconds, so that the fabric can stabilize. At that point, if the Fibre Channel
adapter driver detects that the device is not on the fabric, it begins failing all
I/Os at the adapter driver. Any new I/O or future retries of the failed I/Os are
failed immediately by the adapter until the adapter driver detects that the
device has rejoined the fabric.
Fast Failure of I/O is controlled by a new fscsi device attribute, fc_err_recov.
The default setting for this attribute is delayed_fail, which is the I/O failure
behavior seen in previous versions of AIX.

Note: When you only have one single Fibre Channel interface, it is
recommended to leave the fc_err_recov attribute on delayed_fail.

For more information about dyntrk and fc_err_recov, refer to the following
Web page:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/index.jsp?topic=/c
om.ibm.aix.baseadmn/doc/baseadmndita/dm_mpio.htm&resultof=%2522rmpat
h%2522%2520&searchQuery=rmpath&searchRank=1&pa
› max_xfer_size
This is an AIX setting that can directly affect throughput performance with
large I/O blocksizes. For more information about max_xfer_size refer to IBM
Midrange System Storage Hardware Guide, SG24-7676.
› lg_term_dma
This is an AIX setting that can directly affect throughput performance with
large I/O blocksizes. For more information about lg_term_dma refer to IBM
Midrange System Storage Hardware Guide, SG24-7676.
Chapter 7. Live Partition Mobility 313
› num_cmd_elems
This tunable represents the maximum number of requests that can be
outstanding on or queued to a disk adapter. Change this tunable to adapt to
various memory and system conditions. The default is 200.

Attention:
› There is a relation between the queue_depth tunable on a disk and the
num_cmd_elems on a FC card.
You can tune the num_cmd_elems as follows:
num_cmd_elems = (number of paths that form the vpath) * (value of
queue_depth of the disk)
For example, if you have four paths forming the vpath to your storage and
your queue depth is 20, num_cmd_elems need to be set at 4 * 20 = 80.
› Fast I/O Failure is useful in situations where multipathing software is used.
Setting the fc_err_recov attribute to fast_fail can decrease the I/O fail times
because of link loss between the storage device and switch. This supports
faster failover to alternate paths.
› In single-path configurations, especially configurations with a single path to
a paging device, the delayed_fail default setting is recommended.

For more information about how to set those parameters, refer to Chapter 3,
“Storage layout” on page 85.

Disk mapping versus LUN masking on VIOS

It is important to perform disk mapping on the VIOSs to the client logical partition
for the migration to be successful. LUN masking is not supported. In other words,
the LUN that you create on the SAN storage box, when discovered on the VIOS
with the cfgdev command must be mapped entirely to the mobile logical partition.
314 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Tips:
› Run the cfgdev command on the VIOS as padmin to discover newly added
devices that you want to see after you booted the system.
› When using the mkvdev command to map your disk from the VIOS to the
logical partition as shown in the following example, think of adding the -dev
parameter that enables you to customize the name of your mapping:
For example, mkvdev -vdev hdisk1 -vadapter vhost0 -dev
tmp_swap_mobile
› Both rootvg as data volume groups need to be mapped entirely from the
VIOSs to the logical partitions. This is known as disk mapping. In other
words, there is no need to define any logical volume on the disk to map it to
the client LPAR.

In the example we created a disk mapping on hdisk5, giving it the name

tmp_swap_mobile to the target device. The Universal Device ID (UDID), the
Physical Volume ID (PVID) or the IEEE attributes can be used to verify that your
have the same disk on the source and destination system.

Note: The device cannot have an IEEE volume attribute identifier. Use the
UDID or PVID of the device to identify uniquely your disk on both VIOSs.

Tip: You can establish a correspondence table to describe the LUNs’ IDs on
the storage side compared to the LUNs’ IDs created with the mkvdev command
on the VIOSs. See the following table for an example:

7.4.8 Summary
Power System virtualization offers a set of resource partitioning and
management features such as LPARs, the DLPAR facility, and virtual I/O, under
which you can implement SEA, virtual Ethernet, virtual SCSI, or a VLAN. Shared
Chapter 7. Live Partition Mobility 315
 processor partitions allow you to create an LPAR using as little as 0.10 of a
processor. The DLPAR facility enables you to change the LPAR resources
(processor, memory, and I/O slots) at run time, without rebooting the operating
system.

The versatility of the DB2 data server and a variety of possible combinations of
Power System virtualization features help DB2 applications to perform optimally
in many situations. The Power System virtualization technology can be
configured to realize both computer system and business benefits, which
includes high performance, workload isolation, resource partitioning, maximum
resource use, and high availability at low cost. This technology reduces total cost
of ownership (TCO) while also enhancing expandability, scalability, reliability,
availability, and serviceability.

The best practices presented in this document are essentially lessons that have
already been learned through our own testing. These best practices serve as an
excellent starting point for using the DB2 product with Power System
virtualization. You can use them to help to avoid common mistakes and to
fine-tune your infrastructure to meet your goals for both your business and IT
environment. To validate the applicability of these best practices before using
them in your production environment, establish a baseline and perform sufficient
testing with the various virtualization features.

7.5 Migration process and flow

When you move a logical partition from one system to another, a few of its
attributes might change (such as the logical partition ID number) and a few of its
attributes remain the same (such as the logical partition configuration). You can
choose the way the migration is done by giving a new name to your migrated
partition, changing form VIOS mover, and so forth.

Table 7-11 summarizes those major attributes that might change.

Table 7-11 Logical partition attributes that might change after a migration
LPAR attributes that remain the same LPAR attributes that can be optionally changed
by the user

The logical partition type The logical partition name

› dedicated versus shared processor
› dedicated versus shared memory

The logical partition’s profile The logical partition ID number

316 Best Practices for DB2 on AIX 6.1 for POWER Systems
 LPAR attributes that remain the same LPAR attributes that can be optionally changed
by the user

Simultaneous Multi-Threading (SMT) state of each

processor

The virtual MAC address, IP addresses

The disk mappings to the target devices

The affinity characteristics of the Logical Memory

Block (LMB)

7.5.1 Migration validation

Prior to migrating a logical partition from a source system to a destination
system, you want to validate that the migration process is successful. Perform
this validation using either the HMC GUI or the command line. Both come up with
warnings and errors messages if any relevant problems are detected during the
validation process.

The high level procedure of a migration validation, after all requisites for LPM are
met, are described in the following steps:
1. Both the source and destination systems are capable of migrating and both
are Power6 processor-based, as shown in Figure 7-9.

Figure 7-9 POWER6 systems capable to migrate

2. HMC and remote HMC setup as shown in Figure 7-8 on page 309. (See also
7.4.2, “HMC requirements” on page 285).
Chapter 7. Live Partition Mobility 317
Figure 7-10 Same HMC management (Remote HMC support under conditions)

3. VIOS configured as shown in Figure 7-11.

Figure 7-11 VIOS configured

4. LPAR configured for migration.

5. Valid external storage accessible by the logical partition from both the source
or destination VIOS.
6. Logical partition accesses the SAN disks through virtual SCSI or Fibre
Channel only (or a combination of both). See Figure 7-12 on page 319.
318 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 7-12 All disks must be accessible by both systems

7. Active or inactive migration setup (OS level). See Figure 7-13.

Figure 7-13 Operating System level

8. One or multiple physical IP networks providing communication between the

LPAR and both the source and destination systems.
9. The logical partition uses virtual Ethernet adapters, no Integrated Host
Ethernet Adapter (IHEA) or RMC connectivity. See Figure 7-14 on page 320.
Chapter 7. Live Partition Mobility 319
 Figure 7-14 Network connectivity with all partitions and HMC

7.5.2 Active partition migration

The migration is successful if the validation ended successfully. After all
requirements to allow partition mobility are met and the validation is successful,
you are ready to perform the migration.

The overall migration flow is described in the following process:

1. Run the migration process, either through the HMC GUI or through the
command line.
2. Determine the destination profile name.
3. Specify the remote HMC (see “Dual VIOS” on page 292 for non-symmetric
configurations).
4. Select the destination managed system.
5. Partition validation errors and warnings.
6. Select the MSP.
7. Select the VLAN to be used on the destination system.
8. Select the virtual SCSI adapters to use on the destination system.
9. Select the shared processor pool to be used on the destination system.
10.Select the maximum wait time to migrate (5 minutes is the default).
320 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.5.3 Inactive partition migration
After all requirements to allow partition mobility are met and the validation is
successful, you are ready to perform the migration.

Tip: The major differences in the migration flow between active and inactive
partition migration are:
› The MSP attribute is only mandatory for active partition mobility.
› The wait time can be modified when migrating an active partition.

The overall migration flow is described in the following steps:

1. Run the migration process, either through the HMC GUI or through the
command line.
2. Determine the destination profile name.
3. Specify the remote HMC (see “Dual VIOS” on page 292 for non-symmetric
configurations).
4. Select the destination system.
5. Partition validation errors and warnings.
6. Select the VLAN to be used on the destination system.
7. Select the virtual SCSI adapters to use on the destination system.
8. Select the shared processor pool to be used on the destination system.

To perform the migration, perform the following steps:

1. In the navigation area, select Systems Management and select your inactive
system.
2. Select your mobile partition.
3. Select Operations and Mobility.
4. Click Migrate.

Note: During migration, you can monitor the progress of the migration on both
systems, using IVM or HMC, depending on your environment.
Chapter 7. Live Partition Mobility 321
7.6 Mobility in action
Figure 7-15 shows the setup that was used for DB2 and LPM performance
characterization. Throughout the test, the setup revolved around best practices
to ensure ease in management and higher performance throughput. As
mentioned, preparation for mobility requires careful storage, system and network
planning.

DS5000

Figure 7-15 Configuring JS43 for LPM

An OLTP workload was set up on DB2 9.7 for the LPM experiment. The objective
of this experiment was to demonstrate that DB2 can be moved transparently from
one server to another and that client applications connected to the database
does not experience any downtime. The OLTP workload is a mixture of read-only
and update-intensive transactions that simulate the activities found in complex
OLTP application environments. Storage and Virtualization best practices were
followed to optimize the throughput of the workload.
322 Best Practices for DB2 on AIX 6.1 for POWER Systems
 The remote DB2 client created 50 connections to the database server, each
connection executing a set of transactions. The database server LPAR was
moved from one server to another while the OLTP workload was executing.

Figure 7-16 shows the DB2 environment settings used in the LPM experiment.

Figure 7-16 DB2 environment settings

7.6.1 Configuring the test environment

The first step to configuring the environment was to install a VIOS on each JS43.
This was accomplished by installing a VIOS mksysb image using NIM. The
internal disk within the blade can be used to house the VIOS server or you might
choose to boot the blade with the SAN, as this is also supported. We chose the
internal disk for the blades.

After the VIOS was installed on each blade, we can connect to the Web-based
IVM. This interface provides a HMC-like GUI that allows an administrator to
configure LPARs, virtual network, and virtual storage on the blade and VIOS. As
this is a Web-based tool, you can point your Web browser at the VIOS host
Chapter 7. Live Partition Mobility 323
name, and you are presented with the IVM login page. To log in, use the VIOS
padmin user ID and password.

Before we can test mobility with the JS43, ensure that the environment is
prepared appropriately to support it. Update the firmware levels of the JS43 and
associated components such as the Fibre Channel (FC) adapters. Download the
latest firmware images for the JS43 and the FC adapters from the JS43 support
site. and apply them to each Blade. Install the latest VIOS fixpacks. (Refer to
“CPU and memory requirements” on page 290).

With the correct software and firmware levels installed, prepare the Blade, the
VIOS, and the LPAR for partition mobility.

What follows is a brief checklist of the tasks performed with the IVM:
1. Enter the PowerVM Enterprise Edition APV key on both Blades. This key is
required to enable the mobility feature on the JS43 Blade.
2. Confirm that the memory region size is the same on both Blades. This
information can be found under View/Modify System Properties in the
Memory tab.
3. Configure an SEA on both VIOS. Enable the Host Ethernet Adapter for
Ethernet bridging. This is required for the virtual Ethernet devices to access
the physical Ethernet adapter and the external network. This is performed
under the View/Modify Host Ethernet Adapter, Properties tab. Select Allow
virtual Ethernet bridging. Under View/Modify Virtual Ethernet and the
Virtual Ethernet Bridge tab, and select the physical adapter to be used as the
SEA. A message displays stating that the operation was successful. The SEA
is now configured.
4. Create an LPAR on the source Blade. Select View/Modify Partition 
Create Partition. Enter the LPAR name, memory, and processor
requirements. Ensure that none of the physical HEA ports are selected.
Under Virtual Ethernet, select the SEA to use (for instance, ent0). Under
Storage Type, select Assign existing virtual disks and physical volumes.
Select the SAN disk assigned to the VIOS, which in our environment was the
DS5300 disks.
5. Click Finish to create the LPAR.
The next step is to install AIX. This can be achieved using a NIM mksysb (or
rte) install.
6. With the AIX installation and configuration complete, you can configure DB2.

Note: PowerVM Enterprise Edition is mandatory to enable LPM on your Power

System.
324 Best Practices for DB2 on AIX 6.1 for POWER Systems
 You are now ready to perform a live partition migration. Do one final review and
check. On each VIOS an SEA has been configured. You might use the lsmap
-all net command to confirm the device configuration on both VIOS.

Verify that the same SAN disk can be seen by both VIOS. Using the lspv
command, check that both VIOS have the same PVID associated with the SAN
storage. Confirm that the AIX LPAR is configured with only virtual devices
(meaning no physical adapters, another prerequisite for mobility).

Note: To size CPU for VIOS appropriately, take these into consideration:
› LPM requires additional CPU during LPAR mobility.
› Shared uncapped CPU give you the flexibility to acquire additional CPU
units whenever needed.

7.6.2 Performing the LPM

In this section we discuss the steps for performing the LPM.

Note: As already mentioned we used PowerBlades during our tests. For more
information about Power Systems using HMC, refer to the IBM Redbooks
publication IBM PowerVM Live Partition Mobility, SG24-7460.

Preliminary test
For the purpose of our tests, a single VIOS has been configured, one per Blade,
and one active AIX LPAR running on the first Blade as a VIO client (VIOC). We
are now ready to perform a live partition migration. During the migration, the first
Blade is known as the source system and the second Blade is the destination
system.

The objective is to move the LPAR, from the Blade in the source system to the
Blade in the destination system. At the end of the migration, the AIX LPAR is
running as a VIOC from the destination system on the other physical Blade. DB2
continues to function throughout the entire migration.
Chapter 7. Live Partition Mobility 325
Prior to the migration, run the lsconf command from AIX, and note the system
serial number (see Figure 7-17).

Figure 7-17 lsconf output prior to the migration

During the migration, DB2 jobs are running on the LPAR. Monitor the system
using the topas command and observe that DB2 processes are consuming
processors during the migration.

In the mean time VIOSs are hand-shaking and transmitting information through
the network.
326 Best Practices for DB2 on AIX 6.1 for POWER Systems
All tasks to perform partition mobility are executed from the IVM, on the source
Blade. To start the migration, select the box next to the LPAR and choose
Migrate from the More Tasks drop-down menu. Refer to Figure 7-18.

Figure 7-18 Migration menu

You are presented with a panel to enter the target system details. Enter the
details and then click Validate. Refer to Figure 7-18.
Chapter 7. Live Partition Mobility 327
Migration validation
During the validation phase, several configuration checks are performed. Some
of the checks include:
› Ensuring the target system has sufficient memory and processor resources to
meet the LPAR's current entitlements.
› Checking there are no dedicated physical adapters assigned to the LPAR.
› Verifying that the LPAR does not have any virtual SCSI disks defined as
logical volumes on any VIOS. All virtual SCSI disks must be mapped to whole
LUNs on the SAN.
› RMC connections to the LPAR and the source and target VIOS are
established.
› The partition state is active, meaning Running.
› The LPAR's name is not already in use on the target system.
› A virtual adapter map is generated that maps the source virtual
adapter/devices on to the target VIOS. This map is used during the actual
migration.
328 Best Practices for DB2 on AIX 6.1 for POWER Systems
After the validation completes successfully, a message stating It might be
possible to migrate the this partition ... appears (Figure 7-19). Click Migrate and
the migration to the other Blade begins. Monitor the status of the migration by
clicking Refresh.

Figure 7-19 Migration validation

Chapter 7. Live Partition Mobility 329
On the target Blade, observe that a new LPAR has been created with the same
name as the LPAR on the source Blade. It has a state of Migrating - Running, as
shown in Figure 7-20.

Figure 7-20 Observe the status of the migration on the destination server
330 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.6.3 What happens during the partition migration phase?
During the active migration of the LPAR, state information is transferred from the
source to the target system. This state information includes such things as
partition memory, processor state, virtual adapter state, NVRAM (non-volatile
random access memory), and the LPAR configuration.

The following list details events and actions that occur during the migration:
› A partition shell is created on the target system. This shell partition is used to
reserve the resources required to create the inbound LPAR, or processor
entitlements, memory configuration, and virtual adapter configuration.
› A connection between the source and target systems and their respective
Power Hypervisor is established through a device called the Virtual
Asynchronous Service Interface (VASI) on the VIOS. The source and target
VIOS use this new virtual device to communicate with the Power Hypervisor
to gain access to the LPAR's state and to coordinate the migration. You can
confirm the existence of this device with the lsdev command on the VIOS.

Tip: Use the vasistat command to display the statistics for the VASI device.
Run this command on the source VIOS during the migration. Observe that
Total Bytes to Transfer indicates the size of the memory copy and that Bytes
Left to Transfer indicates how far the transfer has progressed.

› The virtual target devices and virtual SCSI adapters are created on the target
system. Using the lsmap command on the target VIOS before the migration,
notice that there are no virtual SCSI or virtual target device mappings.
Running the same command after the migration shows that the virtual disk
mappings have been created as part of the migration process.
› The LPAR's physical memory pages are copied to the shell LPAR on the
target system. Using the topas command on the source VIOS, you might
observe network traffic on the SEA as a result of the memory copy.
› Because the LPAR is still active, with DB2 still running, its state continues to
change while the memory is copied. Memory pages that are modified during
the transfer are marked as dirty. This process is repeated until the number of
pages marked as dirty is no longer decreasing. At this point, the target system
instructs the Power Hypervisor on the source system to suspend the LPAR.
› The LPAR confirms the suspension by quiescing all its running threads. The
LPAR is now suspended.
› During the LPAR suspension, the source LPAR continues to send partition
state information to the target server. The LPAR is then resumed.
Chapter 7. Live Partition Mobility 331
› The LPAR resumes execution on the target system. If the LPAR requires a
page that has not yet been migrated, then it is demand-paged from the source
system.
› The LPAR recovers its I/O operations. A gratuitous ARP request is sent on all
virtual Ethernet adapters to update the ARP caches on all external switches
and systems in the network. The LPAR is now active again.
› When the target system receives the last dirty page from the source system,
the migration is complete. The period between the suspension and
resumption of the LPAR lasts a few milliseconds, as you can see in
Figure 7-21. In the meantime, the migrating LPAR displays the message
Partition Migration in progress ... as shown in Figure 7-22 on page 333.

Figure 7-21 Suspension: Resumption of the LPAR during Migration

332 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Note: In Figure 7-21 on page 332 we observed a suspension of about 141 ms.

Figure 7-22 Partition migration in progress

With the memory copy complete, the VIOS on the source system removes the
virtual SCSI server adapters associated with the LPAR and removes any device
to LUN mapping that existed previously.

The LPAR is deleted from the source Blade. The LPAR is now in a Running state
on the target Blade. The migration is 100% complete.
Chapter 7. Live Partition Mobility 333
Now that the LPAR is running on the other Blade, run the lsconf command to
confirm that the serial number has changed with the physical hardware. See
Figure 7-23.

Figure 7-23 lsconf output after migration occurred

To confirm and verify that DB2 is not impacted by the migration, check the DB2
alert log for any errors. The ssh login sessions on MOBILE-LPAR remained
active and did not suffer any connectivity issues as a result of the live migration
(see Figure 7-21 on page 332).

Mobility activity is logged on the LPAR and the source and target VIOS. Review
the logs with the errpt (AIX) and errlog (VIOS) commands. On AIX, notice
messages similar to CLIENT_PMIG_STARTED and CLIENT_PMIG_DONE.
Additional information from DRMGR, on AIX is also logged to syslog (for
instance, starting CHECK phase for partition migration). On the VIOS, find
messages relating to the suspension of the LPAR and the migration status (Client
partition suspend issued and Migration completed successfully).

Output of the errpt command from the AIX LPAR and from both VIOSs relate to
the migration activity.
334 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.6.4 Post migration observations
During the migration, we used nmon to capture data that we analyze hereafter.

The migration took about 20 minutes to complete. The LPAR being moved was
configured with 40 GB of memory. Most of the time required for the migration was
for the copying of the LPAR's memory from the source to the target system. The
suspend of the LPAR itself lasted no more than 141 milliseconds. Consider using
a high-performance network between the source and target systems. Also, prior
to the migration, we suggest reducing the LPAR's memory update activity. Taking
these steps improves the overall performance of the migration. (See “Mover
Service Partition” on page 288.)

Note: The time to migrate depends on the quantity of memory allocated to the
logical partition. The more memory allocated, the more time you need to
migrate.

Looking at the error reports show the migration has occurred and reveals the
status of the migration.

LPM has enormous potential for dramatically reducing scheduled downtime for
system maintenance activities. Being able to perform scheduled activities, such
as preventative hardware maintenance or firmware updates, without disruption to
user applications and services is a significant enhancement to any System p
environment. Additionally, this technology can assist in managing workloads
within a System p landscape. It gives administrators the power to adjust resource
usage across an entire farm of System p servers. LPARs can be moved to other
physical servers to help balance workload demands.

Note: For the tests, we used Power Blade systems, each configured with IVM
and one VIOS (one entire CPU, 3 GB of memory). The mobile LPAR had one
entire CPU and 40 GB of memory allocated to it. As such, screen captures
taken from the IVM might differ from screen captures taken with an HMC.
Chapter 7. Live Partition Mobility 335
 OLTP workload
The OLTP workload was running when the migration started. It ran on the source
server during the pre-migration phase. There was a short interval where the
throughput dipped. This was when the hand-over happened. The workload
resumed execution on the target server after the handover. See Figure 7-24.

Figure 7-24 LPM hand over

The following diagrams show you what happened at the moment the real hand
over occurs.

The migration process itself started at 11:20 and finished around 11:39. The load
was run about 11:00 and finished after 12:00. The handover from the source
VIOS to the destination system took place at 11:32. The following processes
were running when the migration command was issued:
› ctrlproc, the mover process
› migrlpar, the migration command that performs the validation and migration of
the logical partition.
› seaproc process, part of the shared Ethernet network and related to network.
› accessproc process.
336 Best Practices for DB2 on AIX 6.1 for POWER Systems
CPU level
In the mobile partition, we see a lessening of the CPU activity at the moment the
hand-over happened from the source to the destination system. See Figure 7-25.

Figure 7-25 Mobile partition: Total CPU

Processes Run Queue slowly increased to attain a maximum value of about 80

and decreased to its nominal value after the hand over finished, as shown in
Figure 7-26.

Figure 7-26 Mobile partition: Processes RunQueue

Chapter 7. Live Partition Mobility 337
The VIOSs in Figure 7-27 and Figure 7-28, reveals that the source mover service
partition uses about 40% more CPU than its entitled capacity during the whole
migration process. The migration process started at 11:20 and finished at 11:39.
In the same way VIOS2 exceeded its entitled capacity by about 30%.

Note: After the migration is finished, you can check the error report for Client
Partition Migration Completed. After you see this in the error report, a few
remaining tasks still need to be finalized.

Figure 7-27 VIOS1: Physical CPU versus EC

Figure 7-28 VIOS2: Physical CPU versus EC

338 Best Practices for DB2 on AIX 6.1 for POWER Systems
Disk level
In Figure 7-29 we clearly see the “gap”, moment corresponding to the point
where the source system releases the disks for the destination system (11:32).

Figure 7-29 Mobile partition: Disk use: Total (KB/second)

Figure 7-30, shows a sharp decline in disk activity (hdisk5), again showing the
time (11:32) when the source system hands over to the destination server.

Figure 7-30 Mobile partition: Disk transfer

Chapter 7. Live Partition Mobility 339
You can see the same in Figure 7-31 at the disk adapter level.

Figure 7-31 Mobile partition: Disk adapter use

This phenomenon is much more clear when looking at the VIOSs.

VIOS1 shows the I/O activity prior the hand-over as shown in Figure 7-32. We
see that after the migration, VIOS1 has nothing more to deal with I/O throughput.

Figure 7-32 VIOS1: Disk adapter use

340 Best Practices for DB2 on AIX 6.1 for POWER Systems
However, after the migration occurred (11:32), VIOS2 takes over all I/O, as
shown in Figure 7-33. From that moment, the mobile partition gets its I/O traffic
through VIOS2 on the destination server.

Figure 7-33 VIOS2: Disk adapter use

Memory level
When looking at memory, we can hardly see a change, as shown in Figure 7-34.
This is as expected, as during the migration the application binds to the
destination system’s memory while its already bound memory is copied over the
network or is freed when unused by the application. Nearly all of the memory is
allocated. We see the load on the system finished at 12:04, which released the
memory while the migration itself finished at 11:39.

Figure 7-34 Mobile partition: Memory use

Chapter 7. Live Partition Mobility 341
Memory usage of each VIOS is not impacted by the migration, as demonstrated
in Figure 7-35 and Figure 7-36. We see later in “Memory level” on page 351 that
the same happens without any load on the system, which correlates our findings.

Recommendation: Consider decreasing network traffic on your network while

the migration is running.

Figure 7-35 VIOS1: Memory use

Figure 7-36 VIOS2: Memory use

342 Best Practices for DB2 on AIX 6.1 for POWER Systems
Network level
The only downtime that occurs during a migration is the network connectivity,
which might be inaccessible for about two seconds. Those two seconds are
negligible for most clients that are connected to the application.

In the example and test we made, we did not notice the loss of connectivity. at
Figure 7-37 does not show a loss, but a drastic decrease in activity during a
fraction of a second followed by noticeable raise in power.

Figure 7-37 Mobile partition: Network I/O

Recommendation: Although performing a LPM, keep the network traffic in

the network adapter as minimal as possible. This ensures the maximum
bandwidth for the VIOS MSP to proceed with the migration.
Chapter 7. Live Partition Mobility 343
We captured the results of a ping command that enables us to evaluate more
precisely the down time of the network connectivity (Figure 7-38). In this
example, the inaccessibility of the network was of about 141 ms.

Figure 7-38 Snapshot of a ping command

344 Best Practices for DB2 on AIX 6.1 for POWER Systems
On VIOS1 we see that it is writing away to the other system at a speed of about
75 MBps until 11:32, moment where the hand-over happened. See Figure 7-39.

Figure 7-39 VIOS1: Network I/O

Similarly on VIOS2, we see a symmetrical read from the source system at

exactly the same speed. See Figure 7-40.

Figure 7-40 VIOS2: Network I/O

LPM: Dry run

To measure the impact of the VIOSs on the mobile partition during a LPM, we ran
a dry run (migration without any particular load), while the partition was only
running AIX. This demonstrates that the VIOSs do not bring any overhead.
Moreover it shows the similarities in the graphs we got.

Compared to the “OLTP workload” on page 336, we observe a similar behavior,

disregard the system is loaded or not.
Chapter 7. Live Partition Mobility 345
CPU level
We observe in the graphs that the source VIOS, which is the initiator of the
migration, has its CPU waving from about 0.1 to 0.25 CPU units. At the moment
the migration process starts (02:46), the CPU of the VIOS initiator (VIOS1)
increases instantly and even goes over its entitled capacity to attain about 125%
of the EC, while the destination VIOS also increases its CPU capacity similarly.
This lasts until the mobile partition is migrated (03:00). This behavior is similar to
the testing we did with a loaded system with the exception of the time until the
hand-over occurs, which is shorter for a non-loaded system than for a loaded
one. However, the global migration time is slightly similar. For more information
refer to “CPU level” on page 337.

Attention: From Figure 7-41 and Figure 7-42 on page 347 we notice the
migration activity to migrate the mobile partition from the source to the
destination server. The migration started at 02:46 and ended at 03:01.
However, there are still tasks to be performed to finalize the migration,
including the copying of remaining memory pages still on the source server.
Refer to 7.5, “Migration process and flow” on page 316 for more information
about the migration processes.

Note: The recommendation to configure the VIOSs is to uncap the CPU. The
weight factor needs to be carefully set to allow the VIOSs to get CPU prior to
any other type of partition.

Remember that the VIOSs are the main virtualization servers that cannot
afford lack of resources, or your LPAR suffers or worse, shuts down.

Figure 7-41 VIOS1: Physical CPU

346 Best Practices for DB2 on AIX 6.1 for POWER Systems
Figure 7-42 VIOS2: Physical CPU

The mobile partition is not influenced because it has nothing to do with the
migration process. Moreover, during this test we did not run any load on it.
System activity here is mainly related to running Java processes. See
Figure 7-43.

Figure 7-43 Mobile LPAR: Physical CPU

Chapter 7. Live Partition Mobility 347
In addition, the average number of kernel threads in the run queues of both
VIOSs show a similar significant increase, as shown in Figure 7-44.

Figure 7-44 VIOSs run queue

348 Best Practices for DB2 on AIX 6.1 for POWER Systems
When the migration completed at 03:01, we observed the run queue of the
mobile partition attained a peak, as shown in Figure 7-45. This behavior is
slightly similar to the one observed during our run with a heavy load on the
system. In that situation, the run queue increased slowly until reaching a peak (at
the migration hand-over time) to decrease in the same manner to its original
value.

Figure 7-45 Mobile partition: Run queue

Disk level
We do not expect any particular disk activity on the mobile partition because
there is no load on it. This is reflected in Figure 7-46.

Figure 7-46 Mobile: Disk use

Chapter 7. Live Partition Mobility 349
In the same way, no disk activity appears on the VIOSs, as the client does not
generate any traffic. This is shown in Figure 7-47 and Figure 7-48.

Figure 7-47 VIOS1: Disk use

Figure 7-48 VIOS2: Disk use

350 Best Practices for DB2 on AIX 6.1 for POWER Systems
Memory level
The observation shows that the VIOSs are only using about 1.6 GB of the total
allocated memory, as shown in Figure 7-49 and Figure 7-50.

This is neither more nor less than the VIOS is actually using when doing nothing.
In other words, the migration does not demand more memory during the transfer.
However, this behavior can be influenced by a heavy network traffic, which
demands memory to handle the network IP packets.

Figure 7-49 VIOS1: Memory use

Figure 7-50 VIOS2: Memory use

Chapter 7. Live Partition Mobility 351
The mobile partition is not affected, so it does not show any particular memory
activity, as shown in Figure 7-51. This is the same result we observed while the
system was loaded. Refer to “Memory level” on page 341 for more details.

Figure 7-51 Mobile partition: Memory use

Network level
The network is obviously the part that reveals what is happening during a
migration. As we have already seen in this chapter, the network is doing the job
of migrating from the source to the destination server.

Figure 7-52 on page 353 and Figure 7-53 on page 353 show that in the window
where the migration from the source to the destination server occurs, the network
reaches 75 MBps. When VIOS1 is pushing information to VIOS2, VIOS2 is
receiving that information. This is why both graphs are symmetrical.
352 Best Practices for DB2 on AIX 6.1 for POWER Systems
Moreover, in Figure 7-52 we see little read activity. Similarly in Figure 7-53 we
see write activity. That activity is like a handshaking between both MSPs in the
migration process to transmit all necessary information to rebuild the LPAR’s
profile, and to copy the memory to the destination system.

Figure 7-52 VIOS1: Network I/O

Figure 7-53 VIOS2: Network I/O

Chapter 7. Live Partition Mobility 353
 However, our mobile partition does not show any particular network activity as
shown in Figure 7-54, because we did not load that system, and because it is not
acting in the migration process itself.

Figure 7-54 Mobile partition: Network I/O

Note: If multiple network interfaces are available on a MSP, it is possible

through the HMC command line to select which IP address the mover uses to
transport client partition data during a LPM event. HMC level V7r3.5.0 or later
is required for this function.

7.7 LPM: Summary of actions

The next figures summarize the necessary actions to undertake, before, during
and after the migration process. These figures are followed by two figures
outlining the major errors you can encounter when using the LPM capabilities of
your servers.
354 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.7.1 LPM: Environment configuration
Figure 7-57 on page 356 shows the operations related with environment
configuration.

Figure 7-55 LPM: Environment configuration

7.7.2 LPM: Pre-migration tests

Figure 7-56 shows the pre migrations tests (validation).

Figure 7-56 LPM: Pre-migration tests (validation)

Chapter 7. Live Partition Mobility 355
7.7.3 LPM: Real migration
Figure 7-57 shows the real migration operations.

Figure 7-57 LPM: Real migration

7.7.4 Post migration

Figure 7-58 shows the post-migration operations.

Figure 7-58 LPM: After migration

356 Best Practices for DB2 on AIX 6.1 for POWER Systems
7.7.5 Problem determination
Figure 7-59 and Figure 7-60 on page 358 show the operations related with
problem determination.

Figure 7-59 LPM: Problem determination, part 1 of 2

Chapter 7. Live Partition Mobility 357
Figure 7-60 LPM: Problem determination, part 2 of 2
358 Best Practices for DB2 on AIX 6.1 for POWER Systems
 8

Chapter 8. Workload partitioning

A Workload Partition (WPAR) is an isolated execution environment with its own
init process inside the AIX environment. It creates a virtualized operating system
environment for managing multiple workloads. To the user, WPAR appears as a
standalone AIX system.

In this chapter we discuss WPAR concepts and the how DB2 can be configured
under a WPAR environment.

This chapter has the following sections:

› “Overview of WPAR” on page 360
› “Other types of WPARs” on page 361
› “Installing and configuring DB2 on a WPAR” on page 364
› “Benefits of WPAR” on page 367
› “Live Application Mobility versus Live Partition Mobility” on page 370
© Copyright IBM Corp. 2010. All rights reserved. 359
8.1 Overview of WPAR
WPAR is the software-based partitioning provided by the operating system.One
AIX operating instance is partitioned into multiple instances, each called a
workload partition. Each workload partition can host applications and isolate
them from applications executing within other WPARS. Figure 8-1 shows a
sample LPAR and WPAR configuration.

Figure 8-1 LPAR and WPAR

Prior to WPARs, it was required to create a new logical partition (LPAR) for each
new isolated environment. With AIX 6.1, this is no longer necessary, as there are
many circumstances when one can get along fine with multiple WPARs within
one LPAR. Why is this important? Every LPAR requires its own operating system
image and a certain number of physical resources. Although you can virtualize
many of these resources, there are still physical resources that must be allocated
to the system. Furthermore, the need to install patches and technology upgrades
to each LPAR is cumbersome. Each LPAR requires its own archiving strategy
and DR strategy. It also takes time to create an LPAR through a Hardware
Management Console (HMC) or the Integrated Virtualization Manager (IVM).
360 Best Practices for DB2 on AIX 6.1 for POWER Systems
 Table 8-1 compares LPARs and WPARs.

Table 8-1 LPAR versus WPAR

LPAR WPAR

Logical partition in a server. Workload partition within LPAR.

Difficult to create and manage (needs HMC, IVM). Easy to create and manage.

Helps to consolidate and virtualize hardware Helps in resource management by sharing AIX
resources in a single server. images and using the hardware resources.

LPAR is the single point of failure for WPAR.If LPAR if WPAR fails then only that instance which was
fails then WPAR fails. partitioned gets affected.

8.2 Other types of WPARs

As mentioned, workload partitions are created within AIX 6.1 instances. The
instance that does not belong to any workload partition is called a global
environment. A system administrator must be logged into the global environment
to create, activate, and manage the workload partitions.

Important: Most performance monitoring and tuning activities are performed

from the global environment.

There are two types of workload partitions that can reside in a global
environment.
› System WPAR: Almost a full AIX environment.
› Application WPAR: Light environment suitable for execution of one or more
processes.
Chapter 8. Workload partitioning 361
 Figure 8-2 shows the global environment and WPAR.

Environment
WPAR1

WPAR2 WPAR4

WPAR3 WPAR5

WPAR4 WPAR6

PPPPPP
System System

Figure 8-2 Global environment and WPAR

8.2.1 System WPAR

System WPAR has the following characteristics:
› Dedicated writable file system. It can share the global environments /usr and
/opt as read only mode or it can has its own private /usr, /opt as well.
› Its own init process
› inetd daemon allows complete networking capacity, so remote login to the
system LPAR is possible
› Runs cron daemon, so execution of process can be scheduled

Note: At the time of writing this IBM Redbooks publication, running NFS
server inside the WPAR was not supported.

Creating a system WPAR

The following sections discuss the creation of a system WPAR with shared /usr
and private /usr.
362 Best Practices for DB2 on AIX 6.1 for POWER Systems
 System WPAR with shared /usr
In this WPAR, /usr and /opt of global environment are shared by the system
WPAR. The following command creates the shared system WPAR:
mkwpar -n <system wpar name>

For example: mkwpar -n sys_wpar_shared_usr.

System WPAR with private /usr

In this WPAR, /usr and /opt are created separately for the system WPAR. The
following command creates the private system WPAR:

mkwpar -l -n “wpar name”

If the network configuration needs to be done while creating the workload, run
the following command:
mkwpar -l -N interface=en0 address="IP" netmask=255.255.255.192 broadcast=9.2.60.255
-n "wpar name having DNS entry"

8.2.2 Application WPAR

An application WPAR has the following characteristics:
› Any application that can be started as a one command in the AIX command
line interface is the candidate for application WPAR.
› It does not own any dedicated storage and it shares the file system of the
global environment.
› It can run daemons but it cannot run any system service daemons such as
inetd, srcmstr, and so forth.
› Remote login is not possible

Creating an application WPAR

The wparexec command is used to create the wpar application as shown:
wparexec -n <“Application wpar name” > <path of the application with
arguments,if any>

For example:
wparexec -n appwpar /tmp/myApp
Chapter 8. Workload partitioning 363
 Example 8-1 shows a sample output of the wparexec command.

Example 8-1 WPAR command

# wparexec /usr/bin/ls
Starting workload partition ls.
Mounting all workload partition file systems.
Loading workload partition.
.com.zerog.registry.xml ibm perf
LicenseUseManagement ifor preserve
aacct lib run
adm locks security
ct log snapp
db2 lost+found spool
ecc msgs ssl
empty ncs statmon
esa news tmp
ha opt websm
hacmp pconsole yp

Shutting down all workload partition processes.

8.3 Installing and configuring DB2 on a WPAR

DB2 installation is supported only on a system WPAR. System WPARs either
share the /usr and /opt directories with the global environment, or have a local
copy of the /usr and /opt directories.
› A DB2 product can be installed in a global environment with a DB2 copy
shared with other system WPARs.
When a DB2 copy is installed in a global environment under either the /usr or
/opt directory, which are shared with system WPARs, those system WPARs
use the shared DB2 copy to set up DB2 instances.
› A DB2 product can be installed in a local file system on a system WPAR.
› Each system WPAR manages its own DB2 instances.
› DB2 instances created on one WPAR, or in a global environment, are not
visible from any other system (system WPAR or global environment).
364 Best Practices for DB2 on AIX 6.1 for POWER Systems
8.3.1 Installing the DB2 copy
Installing a DB2 copy on a system WPAR is similar to any other DB2 product
installation, with the following exceptions. The following items cannot be installed
on a system WPAR:
› IBM Tivoli System Automation for Multiplatforms (SAMP)
› IBM Data Studio Administration Console

Note: UnInstalling the DB2 copy considerations:

› System WPAR is active
› All instances within the system WPAR is dropped or update it to other DB2
copy (using “db2iupdt”)

DB2 instance creation under system WPAR

Perform the following steps to create a DB2 instance under system WPAR.
1. Run the following command:
mkwpar -l -n sys_wpar_db2_private

In this command, sys_wpar_db2_privateis the WPAR name.

2. To start the workload partition, execute the following as root:
startwpar [-v] sys_wpar_db2_private
3. Verify the status of the WPAR using the lswpar command, as shown in
Example 8-2.

Example 8-2 Listing the characteristics of WPAR

# lswpar
Name State Type Hostname Directory
-------------------------------------------------------------
sys_wpar_db2_private A S sys_wpar_db2_private
/wpars/sys_wpar_db2_private
Chapter 8. Workload partitioning 365
4. Log in to the newly created WPAR using the clogin command. See
Example 8-3.
clogin sys_wpar_db2_private

Example 8-3 Log in to system WPAR using “clogin”

# clogin sys_wpar_db2_private
**********************************************************************
*
* Welcome to AIX Version 6.1!
*
* Please see the README file in /usr/lpp/bos for information pertinent to this
release of the AIX Operating System.
*
***********************************************************************

# hostname
sys_wpar_db2_private #
df
Filesystem 512-blocks Free %Use Iused %Iused Mounted on
d
Global 196608 141152 29% 1999 12% /
Global 65536 63776 3% 5 1% /home
Global 4653056 2096048 55% 27254 11% /opt
Global - - - - - /proc
Global 196608 193376 2% 11 1% /tmp
Global 10027008 5658984 44% 42442 7% /usr
Global 262144 123080 54% 4371 24% /var

5. Create groups and users using the commands shown in Example 8-4.

Example 8-4 Create groups and users

mkgroup id=998 db2grp

mkgroup id=999
db2iadm1
useradd -g db2grp -u 200 -md /home/db2inst1 db2inst1
useradd -g db2grp -u 201 -md /home/db2fenc1 db2fenc1
passwd db2inst1 (to set the password)
passwd db2fenc1 (to set the password)
366 Best Practices for DB2 on AIX 6.1 for POWER Systems
 6. Update the /etc/hostsfile with the new host name, as shown in Example 8-5.

Example 8-5 Update the /etc/hosts file

#127.0.0.1 loopback localhost # loopback (lo0)

name/address (comment existing entry)
127.0.0.1 sys_wpar_db2_private.xx.xx.com sys_wpar_db2_private (new

entry)
7. Create the DB2 Instance. See Example 8-6.
db2icrt -p 50000 -s ese -u db2fenc1 db2inst1

Example 8-6 Create the DB2 Instance

# ./db2icrt -p 50000 -s ese -u db2fenc1 db2inst1

DBI1070I Program db2icrt completed successfully.

# su - db2inst1
$ db2start
12/02/2009 12:18:51 0 0 SQL1063N DB2START processing was
successful.
SQL1063N DB2START processing was successful.

Note: DB2 cannot be installed in system WPAR that does not have write
permission to global environment's /usr and /opt.

8.4 Benefits of WPAR

WPAR offers customers the following benefits:
› Hundreds of WPARS can be created, far exceeding the capability of other
partitioning technologies.
› WPAR resource controls enable the over-provisioning of resources. If a
WPAR is below allocated levels, the unused allocation is automatically
available to other WPARS.
› Because there are fewer operating system images when WPARs are used,
there is a reduction in the total amount of AIX system administration and
maintenance tasks.
› There is a reduction in the total amount of system resources needed because
you do not need as much CPU and memory capacity when you have fewer
instances of AIX.

Chapter 8. Workload partitioning 367

 › Even though the OS image used for all WPARs in one LPAR is the same, it is
possible to install versions of the applications used in system WPARs.
› When you run an application in a WPAR, your application is isolated from
other applications, making it easier to monitor and control resources for that
application.
› With WPARs, you can configure unique user and security access roles for
system WPARs.
› No performance cost for using virtual devices.
› WPARS support fast provisioning and fast resource adjustments in response
to normal/unexpected demands. WPARS can be created and resource
controls modified in seconds.
› Enablement of Live Application Mobility for planned outages and workload
distribution.

8.4.1 When to use WPARs

As discussed earlier, WPARs are virtualized operating system environments that
are created within a single AIX image. Although they might be self-contained in
the sense that each WPAR has its own private execution environment with its
own file systems and network addresses, they still run inside the global
environment. The global environment, the actual LPAR, owns all the physical
resources of the logical partition. It is important to note that the global
environment can see all the processes running inside the specific WPARs. The
following list details scenarios that call for the use of WPARs.
› Application and workload isolation
WPARs enable development, test, and production cycles of one workload to
be placed on a single system. Every organization has at least these three
environments for their applications. As a common practice, for each of these
environments you need to create a LPAR because in today’s world of high
availability, it is natural to give each application environment its own home. It
can be a difficult task for a system administrator to maintain all these
environments. This is where the WPAR has the most value. Each of the
environments can be deployed on a separate WPAR within a LPAR, reducing
costs through sharing of hardware resources and software resources (such
as operating systems, databases and other tools).
368 Best Practices for DB2 on AIX 6.1 for POWER Systems
 › An ideal sandbox environment
Creating a separate test environment for system administrators has never
been easier. With WPARs, administrators now have the opportunity to install
new software, test out new patches, install new technology levels and
generally have the flexibility to experiment on the system without any impact
to the other users.
› Troubleshooting
It takes a minimal amount of time to create application WPARs, making it ideal
for quickly troubleshooting or reproducing a customer scenario. As these are
temporary resources, they are destroyed as soon as they end, which
simplifies the manageability of these partitions.

8.4.2 When WAPR might not be the best choice

Here are a few cases when using WAPR might not be the best choice:
› Security
As stated previously, WPAR processes can be seen by the global
environment from the central LPAR. If you cannot compromise on application
security, do not use a WPAR.
› Performance
When doing a performance benchmark on an application running on a WPAR,
remember that the WPARs within the LPAR are sharing system resources, so
race conditions are bound to occur that might change benchmark results.
› Availability
If you are in an environment where it is difficult to bring a system down, it is
important to note that when performing maintenance on an LPAR that every
WPAR defined is affected. At the same time, if there is a system panic and
AIX crashes, every WPAR has now been brought down. From this standpoint,
LPARs without WPARs can provide increased availability across your
environment, albeit at a cost that might be prohibitive.
› Production
If your application demands the granularity and complete OS isolation that
LPARs provide, without having multiple environments to worry about, you
must go in for a LPAR as against a WPAR.
› Physical devices
Physical devices are not supported within a WPAR. Although there is a way to
export devices, this can be a problem for applications that require
non-exportable devices. In this case, they are restricted to running in the
global environment.
Chapter 8. Workload partitioning 369
8.5 Live Application Mobility versus Live Partition
Mobility
Live Application Mobility (LAM) provides the ability to move a running WPAR and
its applications, without moving the AIX instance that contains the WPAR. It is
implemented in software and does not have any dependency on a POWER
processor chip. Any POWER processor-based server running AIX 6.1 supports
workload partitions, whereas Live Partition Mobility provides the ability to move a
running LPAR and all of its running applications from one physical server to
another physical server without disrupting operation of the LPAR. It is only
available on POWER6-processor-based servers. The AIX versions supported
are 5.3 TL7 and 6.1.

You achieve near continuous availability because Live Partition Mobility and LAM
move running workloads between Power Systems to eliminate the need for
planned system downtime. Both LPM and LAM eliminate downtime during
planned outages but cannot increase system availability during unplanned
outages such as a hardware crash. Therefore, it is not a replacement for HACMP.

With Live Partition Mobility, you can move an active or inactive partition from one
physical server to another without the user aware of the change happening. With
LAM, however, you have to check-stop your WPAR so that it restarts on the
destination machine. This might result in brief periods of unresponsiveness.

Table 8-2 lists the differences between Partition Mobility and Application Mobility.

Table 8-2 Basic differences between Partition Mobility and Application Mobility
Type Partition Mobility Application Mobility

OS Linux, AIX 5.3, AIX 6.1 AIX6.1

Hardware POWER6 PowerPC 970, POWER4, 5, 6.

Functionality Server to Server Within the same AIX instance.

Active and inactive partition can Stop the application and restart on the destination
be moved machine.This results in brief periods of
unresponsiveness.

Note: Both application mobility and partition mobility can eliminate downtime
during planned outages but does not provide high availability. They are not a
replacement for HACMP.
370 Best Practices for DB2 on AIX 6.1 for POWER Systems
 A

Appendix A. System health check

This appendix describes the tools used for the DB2 and AIX health check.
© Copyright IBM Corp. 2010. All rights reserved. 371
Why collect ongoing data and data before any major
change?
Over time and through planned and unplanned outages, the system can start
behaving differently. This difference can be a positive or it can have adverse
effects (for example, an increase in memory/paging space, queries that used to
complete in a few seconds taking minutes, increase in I/O on certain disks and
so forth). If information is collected at a point in time before such adverse effects
are noticed and the same information is collected after the adverse effects or
while they are occurring, it makes diagnosing the location of the problem easier.
If, for example you collect access plans of a certain query, before the query used
to take a greater amount of time, you can use theseaccess plans to diagnose
why the query is experiencing a performance degradation.

The same applies for AIX-related information. For example, if information

regarding the system parameters and information about the Logical Volume
Manager (such as mount options and file system layout) is collected before I/O
contention, comparing two sets of relevant identical data aids the problem
determination process.

Why perform ongoing health checks and act on these?

Collecting data is useful for problem determination. However, there are ongoing
health checks that assist in ensuring that the system is performing at an optimal
level. Without regular health checks, and given that business applications seldom
remain constant, it is almost certain that the systems overall performance can be
affected. An example of this at the AIX level can monitoring any hot disks and
time spent in waiting for I/O to complete. Regular checks can allow system
administrators and database administrators to avoid future problems but also
allow the system to perform optimally.
372 Best Practices for DB2 on AIX 6.1 for POWER Systems
DB2 health check
Table A-1 and Table A-2 on page 374 summarize the tools you can use for DB2
health check.

Table A-1 Ongoing data and data before any major change
Task Date Location of
Completed Output

db2support
An all-purpose utility that collects data ranging from ICBM / db cfg, table
space layout to contents of history file information. Additional information
can be found at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.trb.doc/doc/t0020808.html
db2support to gather optimizer data
If, for example you collect access plans of a certain query, before the
query used to take a greater amount of time, you can use these access
plans to diagnose why the query is experiencing a performance
degradation. db2support utility invoked in the optimizer mode can be to
collect such data. Additional information can be found at the following
Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.cmd.doc/doc/r0004503.html
db2fodc - perf
db2fodc is a utility that can be invoked to gather performance-related
data. By saving the DB2 performance-related data when the system is
performing well you can establish a baseline to compare performance
issues if the DB2 performance degrades. Additional information can be
found at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.cmd.doc/doc/r0051934.html
db2cfexp / db2cfimp
These utilities are used to export connectivity configuration information to
an export profile, which can later be imported. Additional information can
be found at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.qb.migration.doc/doc/t0023398.html
Appendix A. System health check 373
Table A-2 Ongoing health checks
Task Date Location of
Completed Output

Taking regular backups

This is vital to any disaster recovery strategy. Whether it be taking db2
backups or other forms of ensuring that the database can be recovered,
this task is of utmost importance. Additional information can be found at
the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.ha.doc/doc/c0005945.html
Ensuring logs are archived successfully
If the database is enabled for rollforward recovery, it is crucial to ensure
that the logs have been archived to a safe location in the event of a failure.
Additional information can be found at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.ha.doc/doc/c0006086.html
Space usage in table spaces
This task has been made easier by the introduction of automatic storage.
There are various health monitoring options that can even pages out if
space available falls below defined thresholds. Additional information can
be found at the following Web pages:
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp
?topic=/com.ibm.db2.luw.admin.dbobj.doc/doc/c0052484.html
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp
?topic=/com.ibm.db2.luw.admin.mon.doc/doc/c0011705.html
reorgchk / reorg
Over time data and indexes in table spaces can become fragmented as
they are inserted, updated, and deleted. To check for this and correct it
use the reorgchk and reorg commands.
Additional information can be found at the following Web pages:
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp
?topic=/com.ibm.db2.luw.admin.mon.doc/doc/c0011705.html
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp
?topic=/com.ibm.db2.luw.admin.perf.doc/doc/c0005406.html
runstats
The optimizer needs to use accurate statistics to ensure that the access
plan chosen for any given query is optimal. It relies on the information
gathered as part of the collection of statistics that is performed through
the runstats command. Additional information can be found at the
following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.cmd.doc/doc/r0001980.html
374 Best Practices for DB2 on AIX 6.1 for POWER Systems
Task Date Location of
Completed Output

Rebinding packages
Applications that issue static SQL statements have their access plan
stored in the system catalogs. If there are changes in data and new
statistics have been gathered to reflect these changes, these packages
that contain the data access methods need to be rebound. Additional
information can be found at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.qb.migration.doc/doc/t0022384.html
Dynamic SQL
Without delving much into tasks involved with monitoring, one monitoring
check is included here. That being the time taken by individual dynamic
SQL statements to identify which SQL statements are consuming the
most CPU and are I/O intensive. Additional information can be found at
the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.mon.doc/doc/r0007635.html
db2advis
The design advisor can assist in the creation of indexes or even MQT's
for given workloads and individual SQL statements. After identifying a
particular SQL statement, the db2advis utility can be used to recommend
various indexes and the performance gain that might result as a result of
creating such recommendations. Additional information can be found at
the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.cmd.doc/doc/r0002452.html
Diagnostic logs
Prudent DBAs not only monitoring the state of their database but also
check their diagnostic logs to find errors that might not be apparent on the
surface. Two such logs to keep checking are the db2diag.log and the
Administration notification log. Additional information can be found at the
following Web pages:
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp
?topic=/com.ibm.db2.luw.admin.ha.doc/doc/c0023140.html
› http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp
?topic=/com.ibm.db2.luw.admin.trb.doc/doc/c0020815.html
db2dart / db2 inspect
To check the structural integrity of the underlying database table space
and containers in which the data, indexes, lobs and so forth reside, use
the db2dart and or the db2 inspect utilities. Additional information can be
found at the following Web page:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/index.jsp?
topic=/com.ibm.db2.luw.admin.trb.doc/doc/c0020763.html
Appendix A. System health check 375
AIX health check
Table A-3 summarizes the tools you can use for AIX health check.

Table A-3 AIX health check tools

Task Date Location of
Completed Output

NMON
nmon (short for Nigel's Monitor) is a popular system monitor tool for the
AIX and Linux operating systems. It provides monitoring information
about the overall health of these operating systems. Information about
nmon can be found from the following Web page:
http://www.ibm.com/developerworks/aix/library/au-analyze_aix/
VMSTAT
This tool is useful for reporting statistics about kernel threads,
virtual memory, disks, and CPU activity. Information about VMSTAT
usage can be found in following Web page:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm
.aix.cmds/doc/aixcmds6/vmstat.htm
IOSTAT
The iostat tool reports CPU statistics and input/output statistics for TTY
=devices, disks, and CD-ROMs. See the following Web page for details
on using IOSTAT:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/index.jsp?top
ic=/com.ibm.aix.cmds/doc/aixcmds3/iostat.htm
LPARSTAT
Use this tool to check the resources for the LPAR on AIX.
It can be used to see the overall CPU usage relative to the shared
pool and to get statistics with regard to the power hypervisor. Information
about LPARSTAT usage can be found in the following Web page:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm
.aix.cmds/doc/aixcmds3/lparstat.htm
PM Services
PM Services can be used to moniitor the overall system vitals of Power
on AIX. It can be used as a useful utility for monitoring the overall health
of a system or multiple systems. Information about PM can be found in
following Web page:
http://www-03.ibm.com/systems/power/support/pm/news.html
376 Best Practices for DB2 on AIX 6.1 for POWER Systems
Task Date Location of
Completed Output

PS
The PS command displays statistics and status information about
processes in the system, including process or thread ID, I/O activity, and
CPU and memory use.The PS command can be used to monitor memory
use by an individual process. For more information, see the following Web
page: http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm
.aix.cmds/doc/aixcmds4/ps.htm

NETSTAT
The netstat command displays information regarding traffic on the
configured network interfaces. For more information, see the following
Web page:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm
.aix.cmds/doc/aixcmds4/netstat.htm
SVMON
The svmon can be used for in-depth analysis of memory usage. It
displays information about the current state of memory. For more
information, see the following Web page:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm
.aix.cmds/doc/aixcmds5/svmon.htm
PRTCONF
Get the basic system configuration. For more information see the
following Web page:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/topic/com.ibm
.aix.cmds/doc/aixcmds4/prtconf.htm
Appendix A. System health check 377
378 Best Practices for DB2 on AIX 6.1 for POWER Systems
Abbreviations and acronyms
AIO asynchronous I/O LPAR logical partition
AMS Active Memory Sharing LRU Least Recently Used
Ack acknowledgement LTG logical track group
CA Configuration Advisor LUN logical unit number
CGTT created global temporary LVM Logical Volume Manager
tables MLS multilevel security
COD Capacity On Demand MPIO multiple path I/O
CTQ Command Tagged Queueing
MTU maximum transmission unit
DBA database administrator NIM Network Installation Manager
DFU Decimal Floating point unit NUMA non-uniform memory access
DGTT declared global temporary OLTP online transaction processing
table
PV physical volume
DLPAR dynamic logical partitioning
RAID Redundant Array of
DMS Database managed space
Independent Disks
DW data warehouse RBAC Role Based Access Control
FC Fibre Channel RID row identifier
GPFS General Parallel File System
SAMP System Automation for
HA high availability Multiplatforms
HADR High Availability Disaster SEA Shared Ethernet Adapter
Recovery SMS system managed space
HBA Host Based Adapter SMT Simultaneous Multi Threading
HMC Hardware Management STMM Self Tuning Memory Manager
Console
TCO total cost of ownership
IBM International Business
Machines Corporation VG volume groups
IOCP I/O completion ports VMM Virtual Memory Manager
ITSO International Technical VMX Vector Multimedia extension
Support Organization WPAR Workload Partition
IVE integrated virtual Ethernet
IVM Integrated Virtualization
Manager
JFS2 Journaled Filesystem
Extended
LBAC Label Based Access Control
© Copyright IBM Corp. 2010. All rights reserved. 379
380 Best Practices for DB2 on AIX 6.1 for POWER Systems
Related publications

The publications listed in this section are considered particularly suitable for a
more detailed discussion of the topics covered in this book.

IBM Redbooks
For information about ordering these publications, see “How to get Redbooks” on
page 382. Note that a few of the documents referenced here might be available
in softcopy only.
› IBM PowerVM Live Partition Mobility, SG24-7460
› PowerVM Virtualization on IBM System p: Introduction and Configuration
Fourth Edition, SG24-7940
› Introduction to the IBM System Storage DS5000 Series, SG24-7676
› DB2 UDB V7.1 Performance Tuning Guide, SG24-6012
› Integrated Virtualization Manager on IBM System p5, REDP-4061
› A Practical Guide for Resource Monitoring and Control (RMC), SG24-6615
› Integrated Virtual Ethernet Adapter Technical Overview and Introduction,
REDP-4340
› WebSphere Application Server V6.1: JMS Problem Determination,
REDP-4330
› IBM DB2 9 on AIX 5L with NFS, iSCSI, and FCP using IBM System Storage
N series, REDP-4250
› IBM System p Advanced POWER Virtualization (PowerVM) Best Practices,
REDP-4194
© Copyright IBM Corp. 2010. All rights reserved. 381
Online resources
These Web sites are also relevant as further information sources:
› AIX 6.1 information center:
http://publib.boulder.ibm.com/infocenter/aix/v6r1/index.jsp
› IBM System p information center:
http://publib16.boulder.ibm.com/pseries/index.htm
› PowerVM information center:
http://www-03.ibm.com/systems/power/software/virtualization/index.html
› Power Systems information center:
http://publib.boulder.ibm.com/infocenter/powersys/v3r1m5/index.jsp
› DB2 information center:
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp
› AIX Commands Reference:
http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic
=/com.ibm.aix.doc/doc/base/commandsreference.htm
› Information about HMC upgrade:
http://www-933.ibm.com/support/fixcentral/
› Information about how to set up SSH keys authentication for HMC setup:
http://publib.boulder.ibm.com/infocenter/powersys/v3r1m5/index.jsp
› Information about how to change your current VIOS level:
http://www14.software.ibm.com/webapp/set2/sas/f/vios/download/home.html
› Location for DB2 and NAS over iSCSI demo:
http://www.ibm.com/partnerworld/wps/servlet/ContentHandler/VPAA-7M59ZR

How to get Redbooks

You can search for, view, or download Redbooks, Redpapers, Technotes, draft
publications and Additional materials, as well as order hardcopy Redbooks
publications, at this Web site:
 ibm.com/redbooks

382 Best Practices for DB2 on AIX 6.1 for POWER Systems
Help from IBM
IBM Support and downloads
ibm.com/support

IBM Global Services

ibm.com/services
Related publications 383
384 Best Practices for DB2 on AIX 6.1 for POWER Systems
Index
autonomics 2
Numerics
64-bit kernel 11
80-20 rule 133 B
BACKUP utilities 16
Barrier synchronization 301
A benchmarking 131
Active Memory Sharing (AMS) 23, 250
Blue Gene 4
active migration 270
built-in autonomics 273
active processors 29
advanced DB2 registry parameters 73
DB2_LARGE_PAGE_MEM 73 C
DB2_RESOURCE_POLICY 73 capped mode 28
DB2MEMDISCLAIM 75 catawarehouse environments 98
advanced virtualization capabilities 23 data compression 99
AIX 13 placing tables into tablespaces 98
AIX 5.3 Technology Level 7 13 chpath 253
AIX configuration 35 client memory 36
tunable parameters 36 clustering administration 19
AIX kernel 13 computational memory 36
AIX kernel extensions 13 Configuration Advisor (CA) 16
AIX NUMA 73 configurational differences 80
AIX Virtual Memory Manager (VMM) 36 AIX 5.3 versus AIX 6.1 80
computational memory 36 CREATE TABLESPACE 89
considerations for DB2 40
DB2 considerations
D
lru_file_repage 42
data stream prefetching 6
maxclient% 41
data warehouse (DW) 88, 105
maxfree 41
database administrator (DBA) 86
maxperm% 42
DATABASE_MEMORY parameter 230
minperm% 42
DataPropagator 19
strict_maxclient 42 DB2 17, 19
free list of memory pages 38 DB2 9.7 Enterprise Server Edition 19
Large page considerations 46 DB2 containers 220
non-computational memory 36 DB2 Database Configuration (DB) parameters 80
page replacement 38 DB2 Database Manager Configuration (DBM) pa-
ALTER DATABASE 89 rameters 80
array 104 DB2 HADR (high availability disaster recovery) 272
array configuration 111 DB2 health center 16
Asynchronous Input Output (AIO) 67 DB2 health check 373
consideration for DB2 67 DB2 overrides 83
legacy AIO 67 DB2 Performance Expert 18
Posix AIO 67 DB2 Performance Optimization Feature 18
automatic storage 88 DB2 pureScale 19
Automatic Variable Page Size 14
© Copyright IBM Corp. 2010. All rights reserved. 385
DB2 Query Patroller 18 fc_err_recov 293, 313
DB2 registry parameters 72 fcstat 119
DB2_LOGGER_NON_BUFFERED_IO 72 Fibre Channel (FC) 100
DB2_PARALLEL_IO 73 adapters configuration 118
DB2_USE_FAST_PREALLOCATION 73 max_xfer_size 118
DB2_USE_IOCP 72 num_cmd_elems 118
DB2 Replication 19 file system 126
DB2 storage design 87 caching 97, 220
DB2 Storage Optimization Feature 17 firmware level 281
DB2 user resource limits (ulimits) 70 Floating-Point Arithmetic Standard 5
DB2 Workload Manager 18 fscsi device attribute 313
db2agntdp 201
db2nodes.cfg 75 G
db2pd 129 general parallel file system (GPFS) 36
DBMS memory set 145 global environment 368
deactivated processors 29 global temporary tables (CGTT) 99
DEADLOCKS 140
Decimal Floating point unit (DFU) 5
declared global temporary tables (DGTT) 99
H
HACMP 370
dedicated LPAR 231
hardware management console (HMC) 24, 250,
design advisor 375
300
disaster recovery solution 272
hash joins 136
disk latency 105
hcheck_interval 293, 310
disk mapping 314
health check 371
down level HMC 277
high availability disaster recovery (HADR) 19
DS family 100
host based adapter (HBA) 101, 246
dual-core processor 4
hot spare drive 103
dynamic logical partitioning (DLPAR) 20, 229
huge pages 301
dynamic processor deallocation 6
Hypervisor 229
dynamic SQ
L snapshot data 141
dynamic SQL 141 I
dyntrk 293, 313 I/O per seconds (IOPS) 104
I/O virtualization 32
IBM AIX V6.1 new features 12
E
continuous availability 13
emulation function 7
concurrent AIX kernel update 13
entitled pool capacity 236
dynamic tracing 13
entstat 300
storage keys 13
errlog 334
manageability 14
errpt 334
automatic variable page size 14
establishing a baseline 131
IBM Systems Director Console for AIX 14
Etherchannel 59
Network Installation Manager Support for
extendvg 123
NFSv4 14
solution performance tuning 14
F security 12
fast I/O failure for Fibre Channel 313 encrypted file system 12
fault tolerant array 103 role based access control (RBAC) 12
FC adapter redundancy 49
386 Best Practices for DB2 on AIX 6.1 for POWER Systems
 secure by default 13 J
support for long pass phrases 13 JFS (journaled file system) 36
trusted AIX 12 JFS2 (Enhanced Journaled File System 2) 11
trusted execution 13 jumbo frames 58
virtualization 12 jumbo sized packets 59
Live Application Mobility 12
IBM DB2 Storage Optimization Feature license 99 L
IBM System Storage DS3000 100 Label Based Access Control (LBAC) 17
IBM System Storage DS5000 100 large objects 17
IBM Systems Director Console for AIX 11 LARGE tablespace 94
IBM TotalStorage DS6000 4 lg_term_dma 313
IBM TotalStorage DS8000 4 link quality 101
IBM WebSphere 234 Linux 269
IEEE 802.1Q VLAN-tagging 255 Linux NUMA 73
ifconfig 55 live application mobility (LAM) 370
inactive migration 270 live partition mobility (LPM) 8, 11, 249, 269
inactive processors 29 active migration 270
INLINE LENGTH 93 active partition migration 320
input and output tunable considerations 63 benefits 272
j2_maxpageReadAhead 64 DB2 migration 273
j2_maxRandomWrite 64 definition 269
j2_minPageReadAhead 64 HMC requirements 285
j2_nPagesPerWriteBehindCluster 66 inactive migration 270
maxpgahead 64 inactive partition migration 321
numfsbufs 65 IVM requirements 287
pv_min_pbuf 66 LPAR requirements 294
sync_release_ilock 66 managed system’s requirements 275
INSTANCE_MEMORY parameter 230 migration process and flow 316
instruction-set architecture (ISA) 3 migration validation 317
integrated virtualization manager (IVM) 24, 250, mobility in action 322
360 network requirements 303
Interface Tool (SMIT) 11 overview 270
Internet 81 storage requirements 306
Internet Control Message Protocol (ICMP) 81 summary 315
inter-partition communication 21 system planning 274
introduction VIOS requirements 287
DB2 15–20 what it is not 272
IBM AIX V6.1 11 why you would use 271
Power Systems 3 Load Lookahead (LLA) 6
PowerVM virtualization 21–22 lock escalations 140
virtualization 6–8 LOCK_ESCALS 140
ioo command 63 LOCK_TIMEOUTS 140
IOSTAT 173 logical drive 114
IP forwarding 81 logical host Ethernet adapter (LHEA) 305
ipintrq overflows 53 logical memory blocksize 284
iSCSI 308 logical track group (LTG) size 123
isolated 7 logical unit number (LUN) 114
isolated environment 7, 360 logical volume (LV) 120
IT staff 19
Index 387
 configuring 125 migration support 276
logical volume manager (LVM) 11, 119 migration-aware application 273
low DB2 bufferpool hit ratio 230 mkps 51
LPAR (logical partition) mksysb 323–324
profile 227 mkvdev 315
throughput 232 mkvg 120
LPAR (logical partition) considerations 225 monitoring 129
capped shared processor LPAR 236 activity level monitoring 155
configuration 228 database object monitoring 159
dedicated LPAR 231 enhancements in DB2 9.7 149
dedicated versus shared-processor partitions in-memory metrics 150
232 monitoring tools for AIX 169
dynamic logical partitioning (DLPAR) 229 new administrative views 160
dynamically altered resources 229 planning and tuning 133
micro-partition 231 scenarios 185
monitoring 238 alleviating bottlenecks during index creation
performance test results 231 199
planning 226 alleviating I/O bottleneck 209
profile 227 Alter Tablespace no file system cache 185
running DB2 234 disk performance bottleneck 221
scalability 232 memory consumed by the AIX filesystem
summary 233 cache 216
throughput 232 tools for DB2 134
uncapped micro-partition 234 Average Bufferpool I/O Response Time 136
uncapped shared processor LPAR 236 Buffer Pool Hit Ratio 135
virtual processors 235 db2top 142
lparstat 238–239 dirty page steals 139
lru_file_repage 37, 42 dirty page threshold 139
lsconf 326 Dynamic SQL Metrics 141
lshmc -V 286 lock metrics 140
lslparmigr 291 no victim buffers available 139
lsmap 325 Package Cache Hit Ratio 140
lspartition 299 page cleaning 138
lspartition -all 299 Prefetch Ratio 139
lsrsrc IBM.ManagementServer 299 Rows Read/Row Selected Ratio 140
lssyscfg 282–283 snapshots 135
LVM configuration 119 sort metrics 140
traces 134
Transaction Log Response Time 138
M mover service partition 288
manipulated routes 81 multi-partition databases 18
manual failover 255 multipath I/O software 262
materialized query table (MQT) 99 multiple shared-processing pool (MSPP) 236
max_coalesce 118 architectural overview 237
max_transfer 117, 311 default shared-processor pool 237
max_xfer_size 313 entitled capacity 236
memory regions 46 maximum pool capacity 236
memory virtualization 30 physical shared-processor pool 236
micro-partitioning 4, 21, 231 reserved pool capacity 236
388 Best Practices for DB2 on AIX 6.1 for POWER Systems
multi-threaded architecture 209 Para virtualization 9
Pareto's Principle 133
PCI error recovery 6
N peak demand 234
netstat 176, 300 performance engineers 170
network file system (NFS) 36 performing ongoing health checks 372
network tunable considerations 51 physical components considerations 101
clean_partial_connection 52 cables and connectors 101
ip6srcrouteforward 53 drives 102
ipignoreredirects 53 Host Based Adapter 101
ipqmaxlen 53 hot spare drive 103
ipsendredirects 54 physical partition size 121
ipsrcrouterecv 54 physical shared processor pool 236
jumbo frames 58 physical volume (PV) 116
maximum transmission unit (MTU) 58 physical volume ID (PVID) 315
rfc1323 55 point-in-time monitoring 134
tcp_nagle_limit 55 poor performance
tcp_nodelayack 56 causes 132
tcp_pmtu_discover 56 application design 132
tcp_recvspace 56 system and database design 132
tcp_sendspace 57 system resource shortages
tcp_tcpsecure 57 CPU 132
udp_pmtu_discover 58 Disk I/O 133
udp_recvspace 58 Memory 132
udp_sendspace 58 Network 133
nmon 179, 335 POWER architecture 3
nmon analyzer 179 Power Architecture® technology 3
no command 51 Power Hypervisor 9, 21, 276, 300
usage 51 Power Hypervisor enablement 277
non-computational memory 36 Power processor technology 4
nondestructive change 231 POWER4 11
not 369 POWER5 11
N-Port Identifier Virtualization (N-PIV) 22, 246 POWER6 11
num_cmd_elems 314 POWER6 processor-based systems 6
numclient 36 POWER6 Storage Keys 11
numperm 36 PowerHA™ pureScale technology 19
NVRAM (non-volatile random access memory) 331 PowerVM 21
PowerVM editions 23
O Enterprise Edition 23
OLTP (online transaction processing) 104 Express Edition 23
online reorganization 19 Standard Edition 23
operating system (OS) 110 PowerVM Enterprise Edition 246
OVERHEAD 96 PowerVM Express 246
PowerVM Standard 246
PowerVM virtualization architecture 24
P PPC970 11
padmin 315 prefetch size 273
page fault 38 private sort heap 140
paging space considerations for DB2 49 probevue 13
Index 389
processor virtualization 26 self-tuning memory feature 16
promiscuous mode 304 self-tuning memory manager (STMM) 20
PS 175 shared 28
pureXML 2 shared CPU mode 28
shared Ethernet adapter (SEA) 22, 32, 255
failover 255
Q shared processor pools 236
queue_depth 116, 311 shared sort usage 205
queue_depth attribute 264 shared-processor pools 22
simultaneous multi-threading (SMT) 30
R SLAs (service level agreements) 8
RAID levels 104–105, 110 Snapshot Monitor, Activity Monitor 129
comparision 110 socket buffer space 82
RAID 0 105 software licensing costs 9
RAID 1 106 source routing 82
RAID 10 109 split the hot spares 103
RAID 5 107 state information 331
RAID 6 108 storage hardware 100
Redbooks Web site 382 storage keys 13
Contact us xxvii support for long pass phrases 13
Redundant Array of Independent Disks (RAID) 100 SVMON 176
redundant array of independent disks (RAID) 100 syncd daemon 66
REGULAR tablespace 93 System i 11
reliability, availability, serviceability (RAS) 6
remote live partition mobility 283 T
REORG 16 table spaces 18
reserve_policy 293, 311 temporary 17
resource monitoring and control (RMC) 298 temporary tables 17
restricted use tunables 40 time reference partition (TRP) 280, 292
rfc1323 82 Tivoli Systems Automation/PowerHigh Availability
RISC System/6000 3 272
role based access control (RBAC) 12 TOPAS 176
rootvg 49 total cost of ownership (TCO) 6, 20
row-level compression 17 TRANSFERRATE 96
RS/6000 3 trusted AIX 12
run a benchmark 131 trusted execution 13
RUNSTATS 16 tuning storage on AIX 115
hdisk tuning 116
S multipath driver 115
SAN tools 248
Sarbanes-Oxley 16 U
schedo 69 uncapped micro-partition 234
scheduler tunable considerations 69 uncapped mode 28
SCSI-compliant logical unit numbers (LUNs) 244 unit of work (UOW) 154
segment size 114 universal device ID (UDID) 315
self tuning memory manager (STMM) 273
self-configuring 16
self-healing 16
390 Best Practices for DB2 on AIX 6.1 for POWER Systems
 virtualized consolidated environment 9
V vmo command
varyonvg 123
usage 38
vasistat 331
VMSTAT 170
vector multimedia extension (VMX) 4–5
volume group (VG) 120
vhost adapter 311
configuring 120
virtual 9, 21
creating 120
virtual Ethernet 32
vpm_xvcpus parameter 235
virtual Fibre Channel (NPIV ) 246
vtscsi device 307
virtual Fibre Channel adapter 246
virtual I/O server (VIOS) 243
Active Memory Sharing 250 W
best practices 261 WebSphere Application Server 185–186
CPU settings 264 wireline telecommunications 4
dual virtual I/O Server 251 workload partition 359–360
live partition mobility 249 workload partition (WPAR) 11
logical partition 246 application isolation 368
network interface backup 254 global environment 368
network virtualization 248 live application mobility versus live partition mo-
EtherChannel 249 bility 370
link aggregation 249 sandbox environment 369
live partition mobility 249 self-contained 368
shared Ethernet adapter (SEA) 249 troubleshooting 369
virtual LAN 248 when not be the best choice 369
networking 262 when to use 368
resilience 250 workload isolation 368
SCSI queue depth 264 world wide port name (WWPN) 246–247
SEA threading 266
sizing 257 X
Virtual Fibre Channel (NPIV) 246 XML documents 17
virtual LAN 248
virtual network redundancy 253
Virtual SCSI 244
virtual SCSI redundancy 252
virtual I/O server (VIOS) media support 287
virtual LAN (VLAN) 7, 248
virtual machine 9
virtual memory manager (VMM) 47
virtual processor (VP) 28–29
virtual SCSI 21
virtual storage 33
virtual tape 7
virtualization engine 4
virtualization types 9
full hardware and firmware embedded virtualiza-
tion 9
full virtualization 9
OS based virtualization 9
para virtualization 9
virtualized 9
Index 391
392 Best Practices for DB2 on AIX 6.1 for POWER Systems
 (0.5” spine)
0.475”<->0.875”
250 <-> 459 pages
 Back cover ®

Best Practices for DB2 on

AIX 6.1 for POWER Systems
®

Explains partitioning This IBM Redbooks publication presents a best practices guide for
and virtualization DB2 and InfoSphere Warehouse performance on a AIX 6L with Power INTERNATIONAL
technologies for Systems virtualization environment. It covers Power hardware TECHNICAL
Power Systems
features such as PowerVM, multi-page support, Reliability, SUPPORT
Availability, and Serviceability (RAS) and how to best exploit them with
DB2 LUW workloads for both transactional and data warehousing
ORGANIZATION
Discusses DB2 systems.
performance The popularity and reach of DB2 and InfoSphere Warehouse has
optimization on grown in recent years. Enterprises are relying more on these products
BUILDING TECHNICAL
for their mission-critical transactional and data warehousing
System p workloads. It is critical that these products be supported by an
INFORMATION BASED ON
adequately planned infrastructure. This publication offers a reference
PRACTICAL EXPERIENCE
Covers OLTP and architecture to build a DB2 solution for transactional and data
data warehouse warehousing workloads using the rich features offered by Power IBM Redbooks are developed by
workloads systems. the IBM International Technical
IBM Power Systems have been leading players in the server industry Support Organization. Experts
for decades. Power Systems provide great performance while from IBM, Customers and
delivering reliability and flexibility to the infrastructure. Partners from around the world
This book presents a reference architecture to build a DB2 solution for create timely technical
transactional and data warehousing workloads using the rich features information based on realistic
offered by Power systems. It aims to demonstrate the benefits DB2 scenarios. Specific
and InfoSphere Warehouse can derive from a Power Systems recommendations are provided
infrastructure and how Power Systems support these products.
to help you implement IT
solutions more effectively in
The book is intended as a guide for a Power Systems specialist to your environment.
understand the DB2 and InfoSphere Warehouse environment and for a
DB2 and InfoSphere Warehouse specialist to understand the facilities
available for Power Systems supporting these products.
For more information:
ibm.com/redbooks

SG24-7821-00 0738434191