Ad
More Related Content
What's hot (20)
Viewers also liked (7)
Ad
Similar to SQL Explore 2012: P&T Part 1 (20)
Ad
More from sqlserver.co.il (20)
Recently uploaded (20)
SQL Explore 2012: P&T Part 1
- 1. Performance Tuning Principles
- 2. Principles of performance tuning • The Goal: • Minimum response time and Maximum throughput • Reduce network traffic, disk I/O, memory usage and CPU time • This is achieved through understanding: • Application requirements • Logical and physical structure of the data • Tradeoffs between conflicting uses of the database. (OLTP vs. DSS) • Start optimizing in early stage of development. It tends to be much harder later.
- 3. Principles of performance tuning • Good design will reduce performance problems: • Architecture • Application & queries • Database • Hardware • Identify the areas that will yield the largest performance boosts • Over widest variety of situations • Focus attention on these areas • Consider Peak load, Not average ones. • Performance issues arise when loads are high
- 4. Measuring Performance and Detecting Bottlenecks
- 5. Measuring Performance • The procedure 1. Create a baseline 2. Find bottlenecks 3. Tune system 4. Compare performance with baseline 5. Repeat
- 6. Measuring Performance • Performance Monitor • SQL Traces & SQL Server Profiler • And derivatives • Extended Events • SQL Server Management Studio • Activity Monitor • Dynamic Management Views • Database Tuning Advisor
- 9. Performance Monitor • Primarily used to discover hardware bottlenecks • Can display values that relate to processor, memory and disk activities • And many more counters… • Tips • Run during heavy load • Run it periodically • Compare with historical data to analyze server activity trends.
- 11. Hardware Counters in Performance Monitor • Processor:% Processor Time • If more than 80-90% for long periods, you might have a CPU bottleneck. • System: Processor Queue Length • Should be less than 4 per processor. • If higher for long periods, you have CPU problem. • Network Interface: Bytes received/sec • Network Interface: Bytes sent/sec • Compare to adapter bandwidth
- 12. Hardware Counters in Performance Monitor • Memory: Memory: Page Faults/sec, pages/sec • monitor disk paging which might cause high disk usage and thus, reduce performance significantly. • SQL Server: Memory Manager: Total Server Memory • ,Target Server Memory • Counters should be identical or close • SQL Server: Buffer Manager: Buffer Cache Hit Ratio • Should be 90% or higher
- 13. Hardware Counters in Performance Monitor • PhysicalDisk: Avg. Disk sec/read, sec/write • Determines how long a read or write operation takes. Should be lower than 20ms (0.02s) at all times • PhysicalDisk: Avg. Disk reads/sec, writes/sec • Determines how many read or write operations occur. • A single 15krpm disk supports ~180 iops • PhysicalDisk: Current Disk Queue Length: • Watch out for high values. • SQL Server: Buffer Manager : Page Reads/sec ,Page Writes/sec • Monitor disk activity caused by SQL server
- 14. SQL Server Counters in Performance Monitor • SQLServer: Access Methods : Full scans/sec • How many tables scans occur per sec • SQLServer: SQL Statistics : Batch Requests/sec • How many batches are executed per sec • Compare to next counter (should be higher) • SQLServer: SQL Statistics : Compilations/sec • How many plans are compiled per sec • Compare to next counter (should be higher) • SQLServer: SQL Statistics : Recompilations/sec • How many plans are recompiled per sec
- 15. SQL Server Counters in Performance Monitor • SQLServer: Locks : Lock requests/sec • SQLServer: Locks : Lock waits/sec • SQLServer: Locks : Average Wait Time (ms) • Lock statistics (requests vs. waits) • And many more • Mirroring • Replication • Service broker • Fulltext Search • SQL Agent • …
- 17. Performance Counters Analysis • Perfmon • PAL/Relog • Console applications for perfmon logs analysis • Use to load performance log into a SQL Server database • Continue analysis with queries & reports (Excel/SSRS) • Use aggregate functions for minimum, maximum and average values. • Investigate work-hours vs. off-hours, weekdays vs. weekends
- 18. Demo: PAL
- 21. SQL Trace • Instruments the server to generate events • Event producer is integrated in SQL server: • Connection Manager • Relational Engine • Storage Engine • Lock and Log Manager • Provides a general purpose trace event model • Producers (SQL Server, Analysis Services) • Consumers (SQL Profiler, SMO Based applications)
- 22. SQL Trace Event Examples • Capture detailed data from: • Connection Manager • Batch, RPC, Connection, Disconnect • Relational Engine • Stored Procedures, SQL statements, Query plans • Lock manager • Locks, deadlock graph, • Auditing • Object access (success/failure) • Object creation • Permissions grant/deny/revoke
- 23. SQL Trace • Powerful (server side) filtering • Thresholds, ranges and equalities • Duration > 1000 • SPID in (10,11,12) (expressed as multiple ==) • Include / exclude • SQL like ‘exec sp_trace%’ • Application name not like ‘%profiler%’ • You may filter by any column • DBID, Reads / Writes, Roll name, Severity…
- 24. SQL Trace • Totally configurable • Which events to trace • Which data to collect (columns) • Pick your destination • Trace File (.TRC) • Server side • Client side (via consumer) • Consumer application (e.g. SQL Profiler) • SQL Server table (via SQL Profiler)
- 25. SQL Trace • Other features: • File roll-over on size limit • Stop traces based on time and/or size • Auto-start traces (with jobs or stored procedures)
- 26. SQL Trace
- 27. SQL Server Profiler • Used to start, stop, view and analyze SQL traces • Adds overhead to the low-overhead trace • Helps discover: • Expensive queries • Very large sort & hash operations • Table searches & updates • Index usage • Missing join predicates • Missing statistics
- 28. SQL Server Profiler • Trace management / analysis client • Input sources: • Real time trace • Trace file/table • Analytical capabilities • Filtering, aggregation and sorting • find the worst queries • counts of events by class • Grouping • By Application name / host name / SPID
- 29. SQL Server Profiler • Debugging & QA • Trace SQL exceptions • Multi-user replay • Use the template “TSQL_Replay” • Display resultsets, errors • Measure playback duration • Allows breakpoints & debugging
- 30. SQL Trace Tips • Create your own templates and share them • Tdf files in .ToolsProfilerTemplatesMicrosoft SQL Server • Be selective – don’t trace everything • The events you would normally trace are RPC:Completed and SQL:Batch Completed • The columns you would usually trace are: TextData, CPU, duration, reads, writes, StartTime, EndTime, • You should also consider adding DatabaseName, HostName, RowCounts • Play with your filters until you get the right balance – enough data to analyze, but not too much data.
- 31. SQL Trace Tips • Use Profiler to define your trace. • Don’t use Profiler to start the trace. • Instead, save trace definition to script and execute the script to get a low-overhead server side script. • Compare SQL traces before & after performance tuning, to show improvement.
- 32. Demo: Trace Best Practices
- 33. SQL Trace Analysis • Use SQL Profiler for line-by-line analysis • To find offending queries & stored procedures: • Load into a table and aggregate by Textdata • Use free tools such as ClearTrace or DBSophic Trace Analyzer • Get a top 10 list of offending queries, handle them one by one, collect another trace and repeat the process until requirements are satisfied
- 36. Dynamic Management Objects • Dynamic management views and functions (DMV) return server state information that can be used to monitor the health of a server instance, diagnose problems, and tune performance. • In General, statistics returned by DMVs is not persisted. • Low overhead – Many DMVs expose information that needs to be maintained anyway
- 37. DMV Categories Category Prefix Common Language Runtime (CLR) Sys.dm_clr_* Database Sys.dm_db_* Indexing Sys.dm_db_index_* Database Mirroring Sys.dm_db_mirroring_* Execution Sys.dm_exec_* Full-Text Search Sys.dm_fts_* I/O Sys.dm_io_* Query Notifications Sys.dm_qn_* Replication Sys.dm_repl_* Service Broker Sys.dm_broker_* SQL Server Operating System Sys.dm_os_* Transactions Sys.dm_tran_*
- 38. DMV Categories Category Prefix Change Data Capture Sys.dm_cdc_* Object Sys.dm_sql_* Resource Governor Sys.dm_resource_governor_* SQL Server Extended Events Sys.dm_xe_* Sys.dm_cryptographic_* Security Sys.dm_provider_* Sys.dm_audit_*
- 39. DMV Examples • Execution plans ‒ sys.dm_exec_query_stats ‒ sys.dm_exec_sql_text ‒ sys.dm_exec_cached_plans • Indexes ‒ sys.dm_db_index_operational_stats ‒ sys.dm_db_index_physical_stats ‒ sys.dm_db_missing_index_details • Database I/O ‒ sys.dm_db_file_space_usage ‒ sys.dm_io_virtual_file_stats
- 40. Demo: DMVs and sys2 Enhancements
- 41. SSMS • Activity Monitor • New features! • Built in reports • Execution plans • Time and IO statistics
- 42. Demo: SSMS Tips&Tricks
- 44. Database Tuning Advisor An automatic tuning tool that copes with: • Retrieval, update & cursor operations • Multiple databases (single server) • Optimizer’s plan choices & cost functions • Single- and multi-column indexes • Indexes for foreign keys • Disk space increase & decrease • “Only add” vs. “replace & add” indexes • Partitioning & Indexed views (Enterprise edition only)
- 45. Database Tuning Advisor • Goal: Determine Optimal Index Requirements based on query statements • Steps: • Analyze your SQL workload – SQL Script file – Trace file • Tune a workload – Runs a sample selection of queries from the workload through the Query Processor with different index combinations and compares query costs • Recommend changes – Adding, dropping or changing indexes – Partitioning • Implement changes / save to script
- 46. Database Tuning Advisor DBA has full control over tuning effort, disk space, etc.
- 47. DTA Tips • Use with caution! • Treat recommendations as such, not all should be implemented. • Creating many indexes isn’t always better. • It’s always easier to add indexes than to remove DTA recommended indexes later.
- 48. Extended Events: Tzahi Hakikat & Keren Bartal
- 49. Introducing SQL Server 2012 Extended Events Enhancements Keren Bartal Tzahi Hakikat 888 holdings
- 50. Agenda • About us • Introduction to Extended Events • Extended Events 2008 • Extended Events Practical Terminology • Extended Events 2012 Enhancements • Summary
- 51. Agenda • About us • Introduction to Extended Events • Extended Events 2008 • Extended Events Practical Terminology • Extended Events 2012 Enhancements • Summary
- 52. About 888 • 888.com is a global online gaming company. • Our purpose is to provide quality entertainment for people who enjoy gambling. • Giving them the opportunity to do so in a safe, fun, fair, regulated and secure environment.
- 54. 888 Database Environment 50 Production Instances 300 Development Instances 400 Databases 250 TB Of Data 24*7 Availability 99.95 Uptime
- 55. Agenda • About us • Introduction to Extended Events • Extended Events 2008 • Extended Events Practical Terminology • Extended Events 2012 Enhancements • Summary
- 56. Extended Events • General event-handling system for windows servers • Used for problem diagnosis and info gathering and auditing • The Extended Events infrastructure supports the correlation of data from SQL Server and OS
- 57. Extended Events • Support 7 different types of targets • Event and consumer agnostic – Any event can be processed by any consumer – New events can be added, immediately useable • Rich predicate system for filtering • Less overhead than server-side trace queues – 10,000 events processed will consume 1% of single 2GHz processor
- 58. Agenda • About us • Introduction to Extended Events • Extended Events 2008 • Extended Events Practical Terminology • Extended Events 2012 Enhancements • Summary
- 59. Extended Events 2008 • Profiler- • • •
- 60. Extended Events 2008 drawbacks • XE required extensive understanding of system catalog views and DMVs • Event Sessions could only be managed through the use of DDL commands • Reading target data requires the use of XQuery
- 61. Extended Event Metadata • Catalog views for defined session info – server_event_sessions – server_event_session_target – server_event_session_fields – server_event_session_actions – server_event_session_events • DMVs for Event System Metadata – dm_xe_package – dm_xe_objects – dm_xe_object_columns – dm_xe_map_values • DMVs for currently active session info – dm_xe_sessions – dm_xe_session_targets – dm_xe_events – dm_xe_event_actions – dm_xe_object_columns
- 62. Demo Capture errors with XE 2008 • Find events and actions • Create a new event session • View the output
- 63. Agenda • About us • Introduction to Extended Events • Extended Events 2008 • Extended Events Practical Terminology • Extended Events 2012 Enhancements • Summary
- 64. Extended Events Objects Module Packages Events Targets Actions Types Predicates Maps
- 65. Packages • Packages are metadata containers • Packages register at module load time • 9 available packages • package0 - XE system objects (default) • sqlserver - SQL Server related objects • sqlos - SQL Server Operating System (SQLOS) related objects • SQL audit uses private XE package
- 66. Events • An event is a well known point in code • Unique schema for each event • Supports optional fields • Events fire synchronously • 264 events in 2008 R2 • 618 events in 2012
- 67. Actions • programmatic response or series of responses to an event • Can be added to any event • Adds data to the event payload • Actions are invoked synchronously • Trigger a memory dump
- 68. Demo Capture errors using the XE UI • Create an event session • Configure action • Watch live data
- 69. Targets • Target is an event consumer – Can be synchronous or asynchronous • Target types – event_file – event_counter – histogram – etw_classic_sync_target – pair_matching – ring_buffer – event_stream
- 70. Demo Monitor locks Present different types of targets • Ring buffer • Event file • Event counter • Histogram • Pair Matching • Etw_classic_sync_target
- 71. Predicates • Predicates are a set of logical rules that are used to evaluate events when they are processed. • Boolean expressions using flexible operators • Event data • Action data • Global State
- 72. Demo Activity Tracking Present different types of Predicates • Event Predicates • Action Predicates • Global Predicates
- 73. Event Session • The materialization of combination of metadata elements of XE architecture • Multiple targets per session • Event can be in many sessions – Actions/Predicates are per event • Event Session can specify what to do if target can't keep up • Event Session defines data retention • Event session can add or remove events on runtime
- 74. Event Session
- 75. Event life cycle Pre-Collect Collection Predicate evaluation Customizable attribute check IsEnabled check Predicate evaluation Event data collected Publish Actions executed Synchronous targets served Event data buffered for asynchronous targets
- 76. Agenda • About us • Introduction to Extended Events • Extended Events 2008 • Extended Events Practical Terminology • Extended Events 2012 Enhancements • Summary
- 77. Extended Events 2012 Enhancements • User Interface – Advanced & Wizard UI for creating and managing – Display & Analysis • Expanded to other systems – Analysis Services, Replication, PDW • Managed code – Powershell object model for runtime and meta data – Reader API for XEL files and near real time stream
- 78. User Interface • Event Session list – Provides a list of Event Sessions • New Session Wizard – Provides a simplified experience for creating an Event Session • Extended Events display – Tabbed windows that display Extended Events trace data
- 79. Demo Capture queries and group by query hash • Grouping • Aggregation • Save XE to a table
- 80. Extended Events Management API • Management API provides the ability to create and modify event sessions • Provides a complete object model for XE usage by managed applications • Provides a XEReader API for reading event files and event streams coming from a running event session on a server
- 81. Agenda • About us • Introduction to Extended Events • Extended Events 2008 • Extended Events Practical Terminology • Extended Events 2012 Enhancements • Summary
- 82. Extended Event Use Cases • Proactive monitoring – Application errors – Errors log – Event grouping • Troubleshooting – Page Split – blocking • Audit – Monitor the access of privileged and non privileged users
- 84. Summary • SQL Server 2012 offers simplified diagnostic tracing with Extended Events – Management Studio integration provides SQL Server Profiler functionality for Extended Events allowing Event Sessions to be created, modified, and scripted – Management API allows managed applications to be developed that leverage Extended Events
Editor's Notes
- #54: Event: sql_statement_completedActions:plan_handle, query_hash, session_id, sql_textTargets: histogram on query_hash, event fileIn target display – choose columns: cpu_time, duration, logical_reads, query_hashGroup by query_hashFrom XE menu – aggregate by cpu_time and durationSave to table
- #60: http://www.sqlserver.co.il/?p=1574
- #61: Demo – extended event 2008 . Create session that record log wait.Start session Read result Stop session -- system catalog?
- #63: Catching errors with profiler: errors and warnings -> user error message -> column filters -> severity >= 16Can’t see the actual query that caused the errorXE allows us to add additional fields like sql_text, using an uncomfortable UI to create the capture and then script to parse the xml data
- #65: Module – exe or dllPackage - a container for XE objects
- #66: package0 - XE system objects (default)sqlserver - SQL Server related objectssqlos - SQL Server Operating System (SQLOS) related objects
- #67: monitoring points of interest the code
- #68: Actions are:programmatic response or series of responses to an eventActions are invoked synchronously on the thread that fired the eventActions can:Capture a stack dump and inspect data.Store state information in a local context using variable storage.Aggregate event data.Append data to event data.Examples of actions:Stack dumperExecution plan detection (SQL Server only)Transact-SQL stack collection (SQL Server only)Run time statistics calculationGather user input on exception
- #69: Create event session Event: error_reportedActions: client_app_name, database_name, session_id, sql_text(no target)
- #71: Events: lock_acquired, lock_releasedActions: client_app_name, database_name, session_id, sql_textTargets:Ring buffer – store data in the memory, cyclic. Watch live data view target (xml)Parse ring buffer XMLEvent file –data is saved in a file, can be archived and reviewed at different timesView targetEvent counter – Counts the number of eventsHistogram – aggregate data on specific event data field or actionFilter on lock_acquiredBase buckets on field mode (type of lock)Pair Matching – pair two related events and output only events that are not matchedBegin with: lock_acquired, end with: lock_releasedActions: session_id, resource_0, resource_1, resource_2Etw_classic_sync_targetLocks types: 0=NULL - Compatible with all other lock modes (LCK_M_NL)1=Schema Stability lock (LCK_M_SCH_S)2=Schema Modification Lock (LCK_M_SCH_M)3=Shared Lock (LCK_M_S)4=Update Lock (LCK_M_U)5=Exclusive Lock (LCK_M_X)6=Intent Shared Lock (LCK_M_IS)7=Intent Update Lock (LCK_M_IU)8=Intent Exclusive Lock (LCK_M_IX)9=Shared with intent to Update (LCK_M_SIU)10=Shared with Intent Exclusive (LCK_M_SIX)11=Update with Intent Exclusive (LCK_M_UIX)12=Bulk Update Lock (LCK_M_BU)13=Key range Shared/Shared (LCK_M_RS_S)14=Key range Shared/Update (LCK_M_RS_U)15=Key Range Insert NULL (LCK_M_RI_NL)16=Key Range Insert Shared (LCK_M_RI_S)17=Key Range Insert Update (LCK_M_RI_U)18=Key Range Insert Exclusive (LCK_M_RI_X)19=Key Range Exclusive Shared (LCK_M_RX_S)20=Key Range Exclusive Update (LCK_M_RX_U)21=Key Range Exclusive Exclusive (LCK_M_RX_X)
- #72: This enables the Extended Events user to selectively capture event data based on specific criteria.New operator – modulus, gt mod, lt mod….
- #73: Fix template…?
- #74: The mapping between package objects and sessions is many to many, which means that that an object can appear in several sessions, and a session can contain several objects.
- #75: The materialization of combination of metadata elements of XE architecture
- #76: Not sure is needed
- #78: UI for creation and management
- #80: Event: sql_statement_completedActions:plan_handle, query_hash, session_id, sql_textTargets: histogram on query_hash, event fileIn target display – choose columns: cpu_time, duration, logical_reads, query_hashGroup by query_hashFrom XE menu – aggregate by cpu_time and durationSave to table
- #84: dbo.trace_xe_event_maphttp://blogs.msdn.com/b/extended_events/archive/2010/12/10/migrating-from-sql-trace-to-extended-events.aspxUSE MASTER; GO USE MASTER; GO SELECT DISTINCT tb.trace_event_id, te.name AS 'Event Class', em.package_name AS 'Package', em.xe_event_name AS 'XEvent Name', tb.trace_column_id, tc.name AS 'SQL Trace Column', am.xe_action_name as 'Extended Events action' FROM (sys.trace_eventsteLEFT OUTER JOIN sys.trace_xe_event_mapem ON te.trace_event_id = em.trace_event_id) LEFT OUTER JOIN sys.trace_event_bindingstb ON em.trace_event_id = tb.trace_event_idLEFT OUTER JOIN sys.trace_columnstc ON tb.trace_column_id = tc.trace_column_idLEFT OUTER JOIN sys.trace_xe_action_map am ON tc.trace_column_id = am.trace_column_id ORDER BY te.name, tc.name