How should I monitor my idaa

A06 - How should I monitor my IDAA?
Mehmet Cuneyt Goksu, mehmet.goksu@ibm.com
IDAA & Db2 Tools Lab Advocate
IBM Germany R&D, Böblingen

Accelerator
Monitoring Interfaces
New: Appliance UI (Accelerator on Z only)
Accelerator
Stored Procedures
SQL
Tools – e.g. OMPE
Reports, Historic Data, Charts
Data Studio
(New: DSM= Data Server Manager)
DISPLAY ACCEL DETAILS

RESOURCES (OBJECTS TO MONITOR)
SQL Statements
Accelerator Resources (Bufferpools, Query Queues, …)
Z Resources (CPU, RAM, Storage, Network)
USE CASES
Capacity
Planning
Performance
Optimization
Accounting
TOOLS
OMPEDSM Own JCLAccelerator Studio Other Vendor tools
SOURCES WITH ACCELERATOR DATA
SMF on z/OS
Db2 Statement
Cache & Explain Tables
Accelerator Status
IFI Stored Procedures Stored Procedures SQL
Query History
in Accelerator
Monitoring Aspects
Workload
Evaluation
System
Utilization

System Utilization, Status and Capacity Planning
• How are system resources utilized over a period of time?
• Is capacity sufficient based on (SLA) models?
• How can I project system resource requirements for higher demands?
• Current system status / health?
USE CASES
Capacity
Planning
Performance
Optimization
Accounting
Workload
Evaluation
System
Utilization
Collecting resource utilization metrics
• Accelerator trace file
• SMF data

System Utilization and Capacity Planning – Tools
DataStudio “Monitor”
DataStudio Query Monitor
• List of latest queries with
execution time and attributes

• Q8STCCPU : The current avg CPU utilization on the accelerator coordinator nodes
• Q8STWCPU: The current avg CPU utilization on the accelerator worker nodes
• Q8STCMEM : Head/Coord node memory utilization
• Q8STWMEM : Average worker nodes memory utilization
• Q8STDSKU : Current disk utilization on worker nodes. The value of this counter is
a percentage (scaled by factor 100) representing disk I/O bandwidth. It is an
average over all workers.
• Q8STABHR : Accelerator Bufferpool Hit Ratio. This is the buffer pool hit ratio for
the accelerator that comprises all currently paired DB2 systems.

System Utilization , Status and Capacity Planning – Tools
• OMPE with Performance Database
• Spreadsheet Generator
• Reports based on SMF counters from Db2 and
Accelerator

Db2 Analytics Accelerator HTAP and V7 Support – Batch
• OMPE 5.4.0
• APAR PI95808, PTF UI56463
• Statistics report & trace
• System parameters report & trace
• IFCID 002 record trace – Db2 Statistics
• IFCID 106 record trace - System Parameters
• PDB and CSV support
• Update and new Tables in T/RKO2SAMP
• Table DB2PMFRTRC_DYNSQL
(eligibility field added)
• New table DB2PMFRTRC_DYNSQL2
• Table DB2PMFRTRC_STASQL
(eligibility field added)
• New table DB2PMFRTRC_STASQL2

Sample Spread Sheet
Running with SQL acceleration
The above graph shows 1 single transaction
execution over 2 hours with the CL2 elapsed (blue)
and CL2 CPU time (red).
At the left it show all 24 txs:
(~2,600 execution / hour)
In the 1st hour 4 out of 24 transactions / SQL were
accelerated. In the 2nd hour none.

IFCID 316 - SQL Statement Statistics – Accelerator data
e.g. Using OMPE Record trace
ACCELERATOR DATA
ACCELERATOR NAME : SIM35
TIME STATEMENT STORED IN CACHE : 11/13/17 16:29:47.365140
STATEMENT IDENTIFIER : 18
MEMBER NAME : 'BLANK'
ACCELERATOR EXECUTIONS : 1 ACCUMULATED # ROWS RETURNED : 7
ACCUMULATED CPU TIME : 0.000001 ACCUMULATED # BYTES RETURNED : 454257
ACCUMULATED ELAPSED TIME : 0.042998 ACCUMULATED EXECUTION TIME : 0.118118
ACCUMULATED QUEUE WAIT TIME: 0.000000 WAIT TIME FOR DB2 : 0.401856
WAIT TIME FOR 1ST ROW : 0.061247
TIME STATEMENT STORED IN CACHE : The date and time when the statement was inserted into the cache
ACCUMULATED CPU TIME : CPU time spent in the accelerator when processing the query request for the statement (reflects parallel processing)
ACCUMULATED ELAPSED TIME : this is the time from the initial request to the last row that is returned to DB2
ACCUMULATED EXECUTION TIME : this is the time spent since starting the query execution until the query execution has finished
WAIT TIME FOR DB2 : time when the first row of the result set was produced by the accelerator until the last row was sent to DB2

IFCID 002 - Accelerator Data - Subsystem/Group
Perspective
SQL STMTS SUCCESSFULLY EXECUTED; The number of SQL statements (sent by this DB2 system since accelerator start) that were successfully executed
CPU TIME IN ACCELERATOR SERVICES; The CPU time of the accelerator services.
CPU TIME FOR REPLICATION; The total CPU cost associated with the replication apply process for this DB2 system.
CPU TIME LOAD/ARCHIVE/RESTORE; The total CPU cost spent in the accelerator for data maintenance operations from this DB2 system

System Status and Health
• What is my system doing?
• E.g. Data Studio
• Is anything misbehaving?
• E.g. OPM System health
or OMPE Exception processing
• Are any queries not running as expected?
• E.g. Query Monitor

Data Studio for IDAA V7
Active queries: The number of accelerated queries
that are currently being processed. In parentheses,
you see the number of queued queries.
CPU Cost
Query execution
CPU time needed for query processing
Data maintenance
CPU time needed for load operations
Replication
CPU time needed for incremental updates

OMPE Real-Time monitoring for IBM DB2 Analytics Accelerator
Accelerator
Statistics
Select and
zoom-in

OMPE Real-Time monitoring for IBM
DB2 Analytics Accelerator
Accelerator
Statistics

Performance Optimization
• How to identify performance problems?
• System and SQL statement performance
• How to tune the system?
• System and SQL statement performance
USE CASES
Capacity
Planning
Performance
Optimization
Accounting
Workload
Evaluation
System
Utilization
Collecting performance and execution details
• Accelerator access plan for individual SQL statements
• SMF data
SQL Statements

Performance Optimization
• Overall system performance optimization
• Bufferpool hit ratio as an indicator for RAM requirements
• Sort overflow as an indicator for RAM requirement
• Query response time classification as an indicator for more system resources (see
‘System utilization’ section)
• SQL performance optimization
• “Explain” statements and tune distribution/organization of data
• Potentially re-write SQL statements with early CTQ operator

Monitoring: DISPLAY ACCEL command for Accelerator V7
Some counters available for Accelerator on PDA are no longer relevant / available in V7
e.g. Maximum queue wait
New for V7
Monitoring: OMPE Statistics for Accelerator V7

System Monitoring – Buffer Pool Hit Ratio
• Single bufferpool for all connected Db2 subsystems
• ‘Total Memory Available for User Data‘ (Q8STTMUD) shows the size
• Buffer Pool Hit Ratio %‘ (Q8STABHR) gives important overall
measure of how effectively the system is using memory to avoid
disk I/O
• Values greater 95% are considered good
• Lower values over a period of time for a given workload may
indicate that the Accelerator is too small to execute the workload
with optimal performance
Db2 Analytics Accelerator

System Monitoring – Sort Overflow
“Total Memory available” (Q8STTMPS) shows memory available for sorts
− Also used for complex query processing, Joins, Group by, …
‘Sort Overflow’ (Q8STOFLW) = Insufficient sort memory for a consumer
− Consumer “spills” results to a system temporary table. May require disk access
− Avoid, as spilling can degrade performance
Many sort overflows over a period of time may indicate that the Accelerator
is too small to execute the workload with optimal performance

DSNX830I -D2UA DSNX8CDA
ACCELERATOR MEMB STATUS REQUESTS ACTV QUED MAXQ
-------------------------------- ---- -------- -------- ---- ---- ----
D2U1IDAA D2UA STARTED 64 0 0 N/A
LOCATION=D2U1IDAA HEALTHY
DETAIL STATISTICS
LEVEL = AQT07017
STATUS = ONLINE
FAILED REQUESTS = 0
AVERAGE QUEUE WAIT = 0 MS
TOTAL NUMBER OF ACTIVE PROCESSORS = 14
AVERAGE CPU UTILIZATION ON COORDINATOR NODES = 8.00%
AVERAGE CPU UTILIZATION ON WORKER NODES = 8.00%
AVERAGE DISK IO UTILIZATION = .00%
NUMBER OF ACTIVE WORKER NODES = 1
TOTAL DISK STORAGE = 4090123 MB
DISK STORAGE IN USE FOR THIS DB2 SYSTEM = 66528 MB
DISK STORAGE IN USE FOR ALL DB2 SYSTEMS = 114528 MB
TOTAL CPU FOR REQUESTS FOR THIS DB2 SYSTEM = 770 MS
TOTAL CPU FOR DATA MAINTENANCE FOR THIS DB2 SYSTEM = 14740 MS
TOTAL CPU FOR REPLICATION FOR THIS DB2 SYSTEM = 5687 MS
TOTAL MEMORY AVAILABLE FOR USER DATA = 113929 MB
TOTAL MEMORY AVAILABLE FOR SQL PROCESSING = 106164 MB
INBOUND NETWORK UTILIZATION OF THE ACCELERATOR = 0 KB
PER SECOND
OUTBOUND NETWORK UTILIZATION OF THE ACCELERATOR = 0 KB
PER SECOND
CURRENT REPLICATION LATENCY FOR THIS DB2 SYSTEM = 5000 MS
NUMBER OF SUCCESSFUL QUERY REQUESTS WITH DELAY PROTOCOL
FOR ALL DB2 SYSTEMS = 0
NUMBER OF EXPIRED QUERY REQUESTS WITH DELAY PROTOCOL
REPLICATION VELOCITY = 10237
Q8STSREQ Q8STNQCS Q8STCQL
Q8STTATE
Q8STFREQ
Q8STQUEW
Q8STCORS
Q8STCCPU
Q8STWCPU
Q8STDSKU
Q8STWNOD
Q8STDSKA
Q8STDSKB
Q8STDSA
Q8STTCQS
Q8STTCMS
Q8STTCCSQ8STTMUD
Q8STTMPS
Q8STANUI
Q8STANUO
Q8STCRL
Q8STCRL
Q8STEDPA
Q8STVLCS

• Q8STSREQ (Requests) : The total number of query/DML Requests sent by this DB2 system that
Successfully executed in IDAA. A request is considered "successful" if no error occurred in the
accelerator backend and also not at the IDAA server. If DB2 does not fetch all query results, the query
execution is still considered as "successful".
• Q8STNQCS (ACTV) : The Number of Requests Currently executing in IDAA on behalf on behalf of this DB2
system. This is a snapshot collected in IDAA. It counts all queries/DML statements know to the (internal)
SQL statement history that are not yet finished. It includes queries in states QUEUED, RUNNING, and
FETCHING as well as DML statements in states QUEUED and RUNNING.
• Q8STCQL(QUED) : The Current Queue Length at the accelerator backend for all DB2 systems. This is a
snapshot, which gets updated for every heartbeat request from DB2.
• Q8STTATE(Status) : Online/Offline/Init/Maintenance/Unknown
• Q8STFREQ (FAILED REQUESTS) : The total number of query Requests sent by this DB2 system to IDAA
and Failed to be successfully executed for any reason. A "failed" request indicates errors in the DB2
client application, DB2, IDAA and/or the accelerator backend. DB2 may generate invalid SQL statements
or route queries/DML that the accelerator backend cannot handle. IDAA may run into internal error
conditions, or the accelerator backend may fail (e.g. due to out-of-memory). Finally, the DB2 client
application may use constructs like division-by-0 or other problems that may cause the query to fail.
• Q8STQUEW(AVERAGE QUEUE WAIT): Current average queue wait time in microseconds. This is the
current average queue wait time. The queue wait time for all queries and DML statements running in the
accelerator backend over the last 1min time interval is averaged.

• Q8STCORS: The total number of logical CPUs cores available on all worker nodes. This
parameter is used as an indicator for the total processing capacity of the accelerator.
The more CPU cores are available, the higher the exploited degree for parallelism during
query processing.
• Q8STCCPU: Current CPU utilization of the Coordinator node. This is a snapshot. The
value is an integer in the range of 0 and 10000, which represents a percentage
multiplied by 100 (to include 2 digits for fractions).
• Q8STWCPU: Current CPU utilization of the Worker nodes.
• Q8STDSKU: Current disk utilization of the worker nodes in percent, i.e. percentage of the
used I/O channels/resources
• Q8STWNOD: Number of accelerator nodes.
• Q8STDSKA : Total disk capacity (for user data) in MB
• Q8STDSKB: Disk space (in MB) for data related to this DB2 system. This value can be set
into relation with Q8STDSKA to determine how much disk capacity has been used.
• Q8STDSA: Disk Space (in MB) for All DB2 systems.

• Q8STTCQS; The Total CPU cost associated with executing Requests (queries/DML) on
IDAA on behalf of this DB2 system. The CPU time includes the CPU time spent in the
accelerator (coordinator and worker nodes) as well as the CPU time spent in IDAA itself.
This value can be put into relation with Q8STTCMS/Q8STTCMA, e.g. to determine the
ratio of data maintenance to data evaluation costs
• Q8STTCMS: The Total CPU cost for data Maintenance operations from this DB2 system.
This counter includes the CPU time for LOAD, ARCHIVE, and RESTORE.
• Q8STTCCS: The Total CPU cost for data Replication
• Q8STTMUD: Total Memory available for User Data in MB. This is not the physical
memory of the accelerator, but the memory configured for bufferpools across all worker
nodes.
• Q8STTMPS: Total Memory available for Processing SQL requests in MB. Memory
available for processing SQL statements in the accelerator
• Q8STANUI: The network usage for the inbound accelerator traffic (current transfer rate
for data received by the accelerator in KB/s).

• Q8STANUO: The network usage for the outbound accelerator traffic (current
transfer rate for data sent by the accelerator in KB/s).
• Q8STCRL: The current replication latency for this DB2 system. Indicator how far
the target accelerator is behind the source. Latency is defined as the time
difference between the timestamp of the last log record was applied to the target
compared to the current time.
• Q8STTDPA: The total number of successful query requests that have been run
under Delay Protocol sent by all db2 systems to the accelerator
• Q8STEDPA: The total number of query requests that have been run under Delay
Protocol sent by all db2 systems to the accelerator that expired because the total
wait time was higher than the specified WAITFORDATA value
• Q8STVLCS: The replication velocity of this db2 system, measured in db2 seconds
replicated per second.

Explaining queries for acceleration from Db2 Analytics Accelerator Studio
For each explained query, a row is inserted into the explain tables:
 PLAN_TABLE and DSN_QUERYINFO_TABLE, if query is accelerated
 PLAN_TABLE's ACCESSTYPE column is set to a value of 'A'
 DSN_QUERYINFO_TABLE's QI_DATA column shows the converted query text
 In DSN_QUERYINFO_TABLE only, if the query is not accelerated
 REASON_CODE and QI_DATA columns provide details
Calls ACCEL_GET_QUERY_EXPLAIN stored procedure
to get an access plan for an accelerated query without executing it

Accelerated query access plan in Db2 Analytics Accelerator Studio

The CTQ Boundary
The CTQ plan operator represents the transition between column-organized data processing and
row-organized data processing
“Keep it columnar as long as possible”
The position of the CTQ and the cardinality flowing out
of the CTQ are indicators for how optimal our query
plan is and how much of the processing is being done in
the column-oriented runtime.
• Reason for early CTQ operator could be non-equi join
of different columns
• Query rewrite might result in an optimized plan
• IBM support (PMR) could suggest rewrite

Workload Evaluation
• How did a certain workload perform?
• Which queries were run?
• Would a certain workload run on the Accelerator?
USE CASES
Capacity
Planning
Performance
Optimization
Accounting
Workload
Evaluation
System
Utilization
Collecting query execution details
• Accelerator query history
• Accelerator / Db2 explain
• SMF data
SQL Statements

Query History (“Query Monitoring” list in DataStudio)
• Stored procedure SYSPROC.ACCEL_GET_QUERIES2
• https://www.ibm.com/support/knowledgecenter/en/SS4LQ8_7.1.0/com.ibm.datatools.aqt.doc/sp_msg/
SPs/sp_idaa_get_queries_2.html
• Extract xml into Db2 table
• http://www-01.ibm.com/support/docview.wss?uid=swg27039739
• Uses ACCEL.GET_QUERY_HISTORY sample (Support doc)
• Based on SYSPROC.ACCEL_GET_QUERIES
• Stored procedure deprecated but functional

Query Monitor, PE Client, …
Analyzing specific workload
• Drill down capability to filter accelerated SQL
executions

Accelerator Modeling
• DB2 estimates the CPU cost and elapsed time for queries eligible for offload to IBM DB2 Analytics Accelerator
• Accelerator Modeling is supported for Accelerator V7
• The availability of DB2 enabling PTFs for Accelerator V7 (DB2 APAR PTF PI95397/UI55280 (Db2 11) including prereq PTFs)
decides whether Modeling is done for Accelerator V5 or Accelerator V7
• PTFs applied: Modeling for V7
• PTFs not applied: Modeling for V5
• General Accelerator Modeling Process:
• Apply or check availability of Accelerator Modeling PTFs (available since 2014)
• PTF UI13488 (Db2 11)
• Set ZPARM ACCELMODEL=YES|NO in DSN6PRM
• For static SQL: Re-bind packages
• Depending on chosen Accelerator Modeling method
• Modeling based on SMF data and Omegamon: Start Accounting Trace and run workload
• Modeling based on dynamic statement cache: Start trace for IFCIDs (316, 317, 318) and run workload
• Based on single dynamic or static statements: Explain statements
• Collect data and evaluate results
• The estimated cost is captured in new fields contained in DB2 accounting record IFCID3
• Supported in both Omegamon Db2 and Query Monitor
• Technote for Statement based Accelerator Modeling:
• http://www-01.ibm.com/support/docview.wss?uid=swg27045618
• ACCELMODEL: https://www.ibm.com/support/knowledgecenter/en/SSEPEK_12.0.0/inst/src/tpc/db2z_ipf_accelmodel.html

OMPE Accounting Report for IBM DB2 Analytics Accelerator
The OMPE Record Trace will show the same details
in its report.
MEASURED/ELIG TIMES APPL (CL1) DB2 (CL2)
------------------ ---------- ----------
ELAPSED TIME 4.830139 4.740227
ELIGIBLE FOR ACCEL N/A 4.442327
CP CPU TIME 6.337894 6.336111
ELIGIBLE FOR SECP 4.990042 N/A
SE CPU TIME 0.000000 0.000000
• regular total class 1 and 2 elapsed times of
the parent only
• Elapsed time spent in DB2 for SQL eligible for
acceleration*. If the statements are executed
in parallel, the elapsed time relates to the
parent task only.
• regular total class 1 and 2 CPU times,
inclusive all parallel threads
• eligible times for specialty engine processing
• The part of CPU time spent on general purpose
processors for SQL eligible for acceleration *.
If the statements are executed in parallel, the
CPU includes the parent and all the
subordinated parallel tasks.
• regular total class 1 and 2 Specialty Engine
CPU times, inclusive all parallel threads
• The part of CPU time spent on specialty engine
processors for SQL eligible for acceleration *.
If the statements are executed in parallel, the
CPU saving includes the parent and all the
subordinated parallel tasks.
-------------------------------------------------
* time can be significantly reduced when SQL is
executed on the accelerator.
Disclaimer: Both the elapsed and CPU time potential saving is the projected upper limit. Namely, even if a
statement is routed to the accelerator there is a DB2 processing associated with preparing and sending
the statement to the accelerator and, particularly, processing associated with receiving the result set and
passing it back to the application. The latter can be significant in case of very large result sets.

Accelerator Modeling - reality check
(produced using OMPE Spreadsheet Input Data generator)
Legend:
• DB2 Class 2 CPU
• Accelerator eligible Class 2 CPU
• Calculated new DB2 CL2 CPU time with acceleration
• Calculated Processing CPU overhead in DB2 during acceleration
Note: In logarithmic scale !!!
„Milage varies“

IBM z Analytics
Monitoring changes in V7

TOTAL MEMORY AVAILABLE FOR USER DATA = 2477 MB
TOTAL MEMORY AVAILABLE FOR SQL PROCESSING = 6243 MB
INBOUND NETWORK UTILIZATION OF THE ACCELERATOR = 1024 KB
PER SECOND
OUTBOUND NETWORK UTILIZATION OF THE ACCELERATOR = 2048 KB
PER SECOND
CURRENT REPLICATION LATENCY FOR THIS DB2 SYSTEM = N/A
NUMBER OF SUCCESSFUL QUERY REQUESTS WITH DELAY PROTOCOL
NUMBER OF EXPIRED QUERY REQUESTS WITH DELAY PROTOCOL
REPLICATION VELOCITY = N/A
DISPLAY ACCEL REPORT COMPLETE
DSN9022I @ DSNX8CMD '-DISPLAY ACCEL' NORMAL COMPLETION
Q8STTMUD
(V7 specific)
Q8STTMPS
(V7 specific)
Q8STANUI
(V7 specific) Q8STANUO
(V7 specific)
Q8STVLCS
(HTAP Specific V5/V7)
Q8STCRL
(HTAP Specific V5/V7) Q8STEDPA
Q8STTDPA
• Q8STANUI : The current inbound transfer rate for data transmitted via network interface between accelerator and DB2 in KB/s.
• Q8STANUO : The current outbound transfer rate for data transmitted via network interface between accelerator and DB2 in KB/s.
https://www-01.ibm.com/support/docview.wss?uid=swg1PI90239

Statistic monitoring metrics
New and deprecated items for V7
Counter Name Description Comment
Q8STTMUD Total Memory available for User Data in MB
(memory configured for buffer pools across all worker nodes)
System Configuration
Q8STTMPS Total Memory available for Processing SQL requests in MB System Configuration
Q8STCQLS Current Queue Length at accelerator backend for this DB2
System
Query Queuing
Q8STTSA Disk space (MB) in use for temporary data - All DB2 subsystems System Utilization
Q8STLSA Disk space (MB) in use for log data (log files) - All DB2 subsystems System Utilization
Q8STOFLW The number of sort overflows in the accelerator backend System Utilization
Q8STABHR Buffer pool hit ratio for the accelerator - All DB2 subsystems System Utilization
Q8STANUI Network usage for inbound accelerator traffic (KB/s received) System Utilization
Q8STANUO Network usage for outbound accelerator traffic (KB/s sent) System Utilization

Stored procedure changes for retrieving queries and
access plans from the Accelerator
• ACCEL_GET_QUERIES (deprecated)
• Supported for V5 compatibility. All SQL statements in the resulting XML string are cut off after the first 128 bytes
• ACCEL_GET_QUERY_DETAILS (deprecated)
• Supported for V5 compatibility only, returns full statement text, but no access plan. AQT20001W instead.
• ACCEL_GET_QUERIES2
• Returns information about past queries and queries that are currently running on an accelerator
• Full SQL statement text - Not truncated after 128bytes
• Returns a result set containing the query information as XML output
• Deprecated ACCEL_GET_QUERIES returned XML output limited to CLOB(4M) parameter
• ACCEL_GET_QUERY_DETAILS2
• Returns information about past queries and queries that are currently running on an accelerator
• Full SQL statement text
• Remote query execution plan in the format of a base64 encoded zip file
• Internal remote query execution plan in text format (db2exfmt, db2look output)
• Returns information in 4 different result sets

SQL-level Monitoring of the Accelerator with
IBM Db2 Query Monitor
IBM Db2 Query Monitor

About Db2 Query Monitor for z/OS
What it is:
• An SQL statement monitor
- Real-time and historical data
- Low overhead
 Traces ACCTG(1,3) and STATS(1,3,4)
 Reads from Db2 control blocks
- Easy to use
 Appropriate for DBAs and application programmers
- A complementary tool to an existing Db2 subsystem monitor
What it isn’t:
• A Db2 subsystem monitor
• A Db2 thread monitor
• A post-processor of trace data
OMEGAMON Db2

Where did my query run?
• Use dynamic statement cache
• EXPLAIN STMTCACHE ALL
• Also supported in Omegamon Db2
• Use static statement cache
• Requires a monitor (no SQL support)
• Supported in OMEGAMON Db2
• Use the accelerator’s query history
• CALL ACCEL_GET_QUERIES

Where did my query run? (continued)
• Use Db2 Query Monitor Both V5 and V7
Accelerators are
supported
• Use drill down capabilities to drill down to SQL statement

Offload Db2 Query Monitor Data to the
Accelerator
• Query Monitor saves data in a VSAM back-store
• Typical retention is 4-15 days
• Data can be loaded into Db2 – and/or the IBM Db2 Analytics Accelerator
• Supports IBM Db2 Analytics Accelerator Loader
• Benefits
• Save data for longer time – even one or more years
• Access historical data using the web UI
• View aggregate data
• Write your own reports – trends & analysis

How to offload
• Create accelerator-compatible tables (ISPF EDIT macro supplied)
• Optionally accelerator-only-tables
• Adapt CQM@LIDA job from sample library (hlq.SCQMSAMP)
• Runs Query Monitor utility that creates sequential datasets for LOAD
• Load both Db2 and accelerator (IDAA_DUAL) or accelerator only (IDAA_ONLY)
• Offload when an interval has been written
• Look for
CQM2401I INTERVAL PROCESSING ENDED FOR QMxx INTV#(00001507)
• Or run daily
Manual: https://www.ibm.com/support/knowledgecenter/SSAURY_3.3.0/topics/cqmucon_loadidaa.html

Viewing Archive Data in Query Monitor Web UI
• You can connect to multiple archives (schemas)
• Archive for specific applications or users
• Archives for daily, weekly and monthly average/totals

AOT Monitoring
• Allows to push down expensive transformation into Accelerator
• Requires V5 or V7 of Accelerator
• IFI Extensions
• IFCID 3/148 (thread level) datasection Q8AC
• IFCID 2 (system statistics) datasection Q8ST

ACCELERATOR DATA - SUBSYSTEM/GROUP PERSPECTIVE
...
| ACCELERATOR SQL CALL DATA
|
|INSERT STMTS SENT TO ACCELERATOR ....: 3 UPDATE STMTS SENT TO ACCELERATOR ....: 7
|DELETE STMTS SENT TO ACCELERATOR ....: 4 DROP STMTS SENT TO ACCELERATOR ......: 8
|CREATE STMTS SENT TO ACCELERATOR ....: 5 COMMIT STMTS SENT TO ACCELERATOR ....: 9
|ROLLBACK STMTS SENT TO ACCELERATOR ..: 6 OPEN STMTS SENT TO ACCELERATOR ......: 10
Batch RECTRACE (IFCID 2) – AOT
Batch RECTRACE (IFCID 2) – AOT

VMNPS08 FOR SUBSYSTEM/GROUP QUANTITY
------------------------------------ --------------------
SQL STMTS SUCCESSFULLY EXECUTED 25.00
SQL STMTS FAILED TO EXECUTE 2.00
CURRENTLY EXECUTING SQL STMTS 0.23
MAXIMUM EXECUTING SQL STMTS 3.00
CPU TIME EXECUTING SQLS STMTS 0.160000
CPU TIME LOAD/ARCHIVE/RESTORE 0.000000
SELECT STMTS SENT TO ACCELERATOR 0.00
INSERT STMTS SENT TO ACCELERATOR 9.00
UPDATE STMTS SENT TO ACCELERATOR 3.00
DELETE STMTS SENT TO ACCELERATOR 3.00
OPEN STMTS SENT TO ACCELERATOR 12.00
CREATE STMTS SENT TO ACCELERATOR 2.00
DROP STMTS SENT TO ACCELERATOR 1.00
COMMIT STMTS SENT TO ACCELERATOR 9.00
ROLLBACK STMTS SENT TO ACCELERATOR 2.00
Batch STATISTICS - AOT

• The accelerator statistics are reported in Q8ST section in IFCID 2
• It is cut at statistical interval (every 1 minute in v10+)
• Q8ST contains data that is accumulated in DB2 and some that is aggregated and returned by the accelerator
via the heartbeat connection
• Q8ST Fields Tracked Within the Accelerator
• These stats are returned to DB2 via the heartbeat connection
• DB2 gets the updated stats every 20 seconds
• Collected since last 'backend' accelerator started, not tied with DB2 startup
• Reset when the 'backend' accelerator is restarted, not tied with DB2 restart
• Some reflect all activity on the accelerator, not just the activity initiated from this DB2
Acceleration Analysis via Statistics Data

© 2018 IBM Corporation
IBM Analytics
53
New and revised counters in Q8ST (Statistics)
Name Description Category New or
Existing
Q8STPRID Accelerator Product Identifier
Problem Determination Existing
Q8STCONN Number of connection to accelerator
Q8STREQ Number of DRDA requests to accelerator
Q8STTOUT Number of timed out requests
Q8STFAIL Number of failed requests
Q8STBYTS Number of bytes sent to accelerator
Q8STBYTR Number of bytes received from accelerator
Q8STMSGS Number of DRDA messages sent to accelerator
Q8STMSGR Number of DRDA messages received from
accelerator
Q8STBLKS Number of DRDA blocks sent to accelerator
Q8STBLKR Number of DRDA blocks received from
accelerator
Q8STROWS Number of rows sent to accelerator
Q8STROWR Number of rows received from accelerator

IBM Analytics
54
Existing
Q8STTCPU The CPU time spent in DB2 to construct the
request to be sent and parse the reply received
via TCPIP
Query Execution
Existing
Q8STTELA The TCPIP elapsed time waited in DB2 to send
the request and receive the result back.
Q8STACPU The CPU time spent in an accelerator
processing the request
Q8STAELA The elapsed time spent in an accelerator
processing the request
Q8STAWAT The time that the query has spent in the
accelerator queues, waiting to be processed
Q8STSREQ Number of successful queries for this DB2
system
Existing
Q8STNQSA Number of successful queries for all DB2
systems
New
Q8STFREQ Number of failed queries for this DB2 system Existing
Q8STNQFA Number of failed queries for all DB2 systems New
Q8STNQCS Number of queries currently (actively) executing
in an accelerator for this DB2 system
New
Q8STACTV Number of queries currently executing in an
accelerator from all DB2 systems
Existing
(8 bytes)
Q8STMNQS Maximum number of queries executing in an
accelerator concurrently on this DB2 system
Existing
(8 bytes)

IBM Analytics
55
Existing
Q8STMAXA Maximum number of queries actively executing
in an accelerator concurrently at any time on
behalf of all DB2 systems
Query Execution Existing
Q8STTCQS The total CPU cost associated with executing
offloaded requests in the accelerator on behalf of
this DB2 system
CPU Utilization
New
Q8STTCQA The total CPU cost associated with executing
offloaded requests in the accelerator on behalf of
all DB2 systems
New
Q8STTCMS The total CPU cost for data maintenance
operations from this DB2 system
New
Q8STTCMA The total CPU cost for data maintenance
operations from all DB2 systems
New
Q8STCCPU Current CPU utilization on coordinator nodes Existing
Q8STWCPU Current CPU utilization on worker nodes Existing
Q8STTCCS The total CPU cost associated to the replication
apply process for this DB2 system
New
Q8STTCCA The total CPU cost associated to the replication
apply process for all DB2 systems
New
Q8STDSKA Total disk capacity in MB
Disk Utilization Existing
Q8STDSKU Current disk utilization for worker nodes in
percentage
Existing

IBM Analytics
56
Existing
Q8STDSKB Disk space in MB for databases for this DB2
system
Disk Utilization Existing
Q8STDSA Disk space in MB for all DB2 systems New
Q8STQUEW Current average queue wait time
Capacity Planning
Existing
Q8STQUEM Maximum queue wait time Existing
Q8STACORS The total number of CPU cores available on all
worker nodes
Existing
Q8STNMDS Degree of parallel I/O channels (also known as
number of data slices)
Existing
Q8STMAXQ Maximum queue length Existing
Q8STWNOD Number of active worker nodes Existing
Q8STCQL Current Queue Length New
Q8STSTATE Accelerator State 0 – INITIALIZING
1 – ONLINE
3 – OFFLINE
5 – MAINTENANCE
255 – UNKNOWN
Existing
Q8STTCSS Replicator State 0 – STARTED
1 – STOPPED
2 – ERROR
3 – STARTING
4 – STOPPING
255 – UNKNOWN
New

IBM Analytics
57
Existing
Q8STTLSC Timestamp when last status change of the
replication state occurred
New
Q8STTART Timestamp when accelerator server process last
started.
Q8STTATC Timestamp when last status change of the
accelerator occurred.
Q8STNLRS Total number of log records read by the
replication capture agent for this DB2 system
Incremental Update via
CDC
Q8STNLRA Total number of log records read by the
replication capture agent for all DB2 systems
Q8STNLTS The number of log records processed by the
replication capture agent for this DB2 system
that are applicable to tables in this accelerator.
Q8STNLTA The number of log records processed by the
replication capture agent for all DB2 systems that
are applicable to tables in this accelerator.
Q8STNBS The number of bytes processed by the capture
agent for this DB2 system
Q8STNBA The number of bytes processed by the capture
agent for all DB2 systems

IBM Analytics
58
Existing
Q8STNIS The number of inserted rows applicable to the
accelerated tables that were processed by the
capture agent for this DB2 system
Incremental Update via
CDC New
Q8STNIA The number of inserted rows applicable to the
capture agent for all DB2 systems
Q8STNUS The number of updated rows applicable to the
Q8STNUA The number of updated rows applicable to the
Q8STNDS The number of deleted rows applicable to the
Q8STNDA The number of deleted rows applicable to the
Q8STCRL The current replication latency for this DB2
system

Acceleration Analysis via Statistics Data
Capacity Planning
• Probably the most used aspect of statistics
• Queuing times + CPU utilization
Query Execution
• Total number of successful queries sent by this DB2 system
• Total number of failed queries sent by this DB2 system
• Number of currently executing queries sent by all DB2 systems
• Max number of concurrently executing queries (high water counter) sent by this
DB2 system
Disk Utilization
• Total disk capacity in MB
• Current disk utilization for worker nodes in percentage

Db2 Analytics Accelerator HTAP and V7 Support – Online
• OMPE 5.4.0
• APAR PH00803, PTF UI58486 (September 2018)
• New fields do not show for IDAA V4
• PE Client APAR PH01934, PTF UI58847

OMPE
Statistics
Report for IBM
Db2 Analytics
Accelerator V5
(PTF 6) & V7
TF3 FOR SUBSYSTEM/GROUP QUANTITY TF3 TOTAL ACCELERATOR QUANTITY
------------------------------------ -------------------- ------------------------------------ --------------------
SQL STMTS SUCCESSFULLY EXECUTED 2.00 SQL STMTS SUCCESSFULLY EXECUTED 3.00
SQL STMTS FAILED TO EXECUTE 0.00 SQL STMTS FAILED TO EXECUTE 0.00
CURRENTLY EXECUTING SQL STMTS 1.00 CURRENTLY EXECUTING SQL STMTS 1.00
MAXIMUM EXECUTING SQL STMTS 1.00 MAXIMUM EXECUTING SQL STMTS 1.00
CPU TIME EXECUTING SQL STMTS 1.790000 CPU TIME EXECUTING SQL STMTS 3.810000
CPU TIME LOAD/ARCHIVE/RESTORE 20.260000 CPU TIME LOAD/ARCHIVE/RESTORE 41.560000
CURRENT QUEUE LENGTH 1.00 DISK STORAGE AVAILABLE (MB) 8024544.00
INSERT STMTS SENT TO ACCELERATOR 0.00 IN USE FOR ACCEL DB - ALL DB2 (MB) 15790.80
UPDATE STMTS SENT TO ACCELERATOR 0.00 IN USE FOR ACCEL DB - THIS DB2(MB) 992.80
DELETE STMTS SENT TO ACCELERATOR 20453.00 IN USE FOR TEMP DATA - ALL DB2(MB) 583.42
OPEN STMTS SENT TO ACCELERATOR 33564.00 IN USE FOR LOG DATA - ALL DB2(MB) 51.61
CREATE STMTS SENT TO ACCELERATOR 2272.00 MAXIMUM QUEUE LENGTH 0.00
DROP STMTS SENT TO ACCELERATOR 2272.00 CURRENT QUEUE LENGTH 0.00
COMMIT STMTS SENT TO ACCELERATOR 0.00 AVG QUEUE WAIT ELAPSED TIME 0.025600
ROLLBACK STMTS SENT TO ACCELERATOR 4.00 MAX QUEUE WAIT ELAPSED TIME 0.464235
CONNECTS TO ACCELERATOR 2.00 MEMORY AVAILABLE - USER DATA (MB) 1048576
REQUESTS SENT TO ACCELERATOR 2254.00 MEMORY AVAILABLE - USER REQUESTS (MB) 524288
TIMED OUT 0.00 SORT OVERFLOWS 0
FAILED 0.00 BUFFER POOL HIT RATIO (%) 98.17
BYTES SENT TO ACCELERATOR 1204453.00 TRANSFER RATE - INBOUND (KB/S) 130318
BYTES RECEIVED FROM ACCELERATOR 4373564.00 TRANSFER RATE - OUTBOUND(KB/S) 216535
MESSAGES SENT TO ACCELERATOR 2272.00
MESSAGES RECEIVED FROM ACCEL 2272.00 WORKER NODES 3.00
BLOCKS SENT TO ACCELERATOR 0.00 WORKER NODES DISK UTILIZATION (%) 0.00
BLOCKS RECEIVED FROM ACCELERATOR 2250.00 WORKER NODES AVG CPU UTILIZATION (%) 2.09
ROWS SENT TO ACCELERATOR 0.00 COORDINATOR CPU UTILIZATION (%) 11.59
ROWS RECEIVED FROM ACCELERATOR 0.00 PROCESSORS 48.00
TCP/IP SERVICES ELAPSED TIME 12.937895 DATA SLICES 22.00
ELAPSED TIME IN ACCELERATOR 0.000000
WAIT TIME IN ACCELERATOR 0.000000 CPU TIME FOR REPLICATION 0.148773
LOG RECORDS READ 137055.00
CPU TIME FOR REPLICATION 0.085255 LOG RECORDS FOR ACCEL TABLES 1456.00
LOG RECORDS READ 945.00 LOG RECORD BYTES PROCESSED 97265.00
LOG RECORDS FOR ACCEL TABLES 940.00 INSERT ROWS FOR ACCEL TABLES 0.00
LOG RECORD BYTES PROCESSED 63209.00 UPDATE ROWS FOR ACCEL TABLES 0.00
INSERT ROWS FOR ACCEL TABLES 0.00 DELETE ROWS FOR ACCEL TABLES 0.00
UPDATE ROWS FOR ACCEL TABLES 0.00
DELETE ROWS FOR ACCEL TABLES 0.00
REPLICATION LATENCY 0.000000
REPLICATON VELOCITY (LOG SECS/SEC) 12.383628 ACCELERATOR SERVER START 12/16/17 09:38:08.97
REPLICATION STATUS CHANGE 12/16/17 09:38:49.79 ACCELERATOR STATUS CHANGE 12/16/17 09:38:14.66
WAITFORDATA - SUCCESSFUL 1.00 WAITFORDATA - SUCCESSFUL 3.00
WAITFORDATA - TIMED OUT 0.00 WAITFORDATA - TIMED OUT 1.00
Data related to Data Replication (CDC)
Capacity planning
Workload Balancing
H
T
A
P

Monitoring HTAP
• IFCID 2 SMF records have new fields:
• Query requests, Query requests expired, Replication velocity
• -DIS ACCEL DETAIL includes new monitoring counters:
CURRENT REPLICATION LATENCY FOR THIS Db2 SYSTEM = 2000 MS
NUMBER OF SUCCESSFUL QUERY REQUESTS WITH DELAY PROTOCOL FOR ALL Db2 SYSTEMS = 270
NUMBER OF EXPIRED QUERY REQUESTS WITH DELAY PROTOCOL FOR ALL Db2 SYSTEMS = 15
REPLICATION VELOCITY (Db2 LOG SECONDS APPLIED PER SECOND) = 1
• ACCEL_GET_QUERIES shows a delayed query as “QUEUED”
• Shown in IDAA Studio, query history section, as query state.
Db2 11 PTF UI51280 with Db2 APARs PI83286 and
PI83288 introduce monitoring counters

How should I monitor my idaa

Recommended

More Related Content

What's hot (20)

Similar to How should I monitor my idaa (20)

More from Cuneyt Goksu (20)

Recently uploaded (20)

How should I monitor my idaa