DB2 Through My Eyes
0 likes566 views
This document discusses techniques for understanding a customer's DB2 environment using readily available system data before speaking with DB2 specialists. It covers analyzing CPU usage, memory usage, I/O, coupling facility usage, XCF traffic, stored procedures, applications, workload manager configuration, DDF rules, and restart patterns using SMF records and other data to detect issues and understand normal behavior. The goal is to "bridge the gap in perspectives between DB2 and system performance specialists."
DB2 Through My Eyes
- 1. Session zPE2235 : DB2 Through My Eyes Martin Packer martin_packer@uk.ibm.com
- 2. Abstract Bridging the gap in perspectives between DB2 and System Performance specialists is a perennial concern of mine: As a specialist in one you're much more valuable if you can bridge that gap. This presentation shows some techniques I use to understand a customer's DB2 environment before I talk to a customer's DB2 specialists (or indeed my own). All these techniques are available to you, the data being readily available. I hope you find them useful.
- 4. What Is DB2? (for the purposes of this discussion)
- 5. Subsystems
- 6. CPU
- 7. CPU ... ā Track CPU by e.g. Time Of Day ā Regular variations ā Spikes ā Detect split between e.g. zIIP-eligible and non- ā For example, Prefetch and Deferred Write might fall away at certain times ā What is a reasonable expectation? ā Notice when zIIP-eligible work ends up on GCPs ā For DIST detect when non-Enclave work is large ā Might suggest issues with e.g. Authentication
- 8. Ā© 2014 IBM Corporation8 A DB2 Version DBM1 Address Space's zIIP and CPU Usage Over 24 Hours
- 9. Real Memory ā SMF 30 data useless for real memory ā Memory service declines if CPU queuing experienced ā Use SMF 72 Subtype 3 ā Reporting Class for individual or groups of DB2 subsystems ā Vast majority will be DBM1 ā Use DB2 IFCID 225 data ā Accurate real memory statistics
- 10. Virtual Memory ā SMF 30 shows allocated ā Above and below the line ā High and Low ā 64-Bit ā Accurate - so far as it goes - for all address spaces ā Use DB2 IFCID 225 data ā Accurate virtual memory statistics for DBM1 ā IRLM, MSTR, DIST much smaller ā Use DB2 Statistics Trace for area usage / value
- 11. I/O ā Both āsystemā and ādatabaseā I/O best represented by SMF 42-6 ā Database and Table Space / Index Space encoded in data set qualifiers ā Relatable to buffer pool by the DB2 Catalog ā Looking for: ā I/O rate ā cacheability* ā Read / Write ratio ā Components of response time * Caution over Sync Remote Copy Disconnect Time
- 13. Parallel Sysplex ā Coupling Facility ā 3 types of Coupling Facility structure: ā GBPs, LOCK1, SCA ā Often duplexed ā Configurations vary considerably ā Lots of useful information ā e.g. LOCK1 False Contention ā e.g. Group Buffer Pool sizing & Data / Directory split ā Augment with DB2 Statistics Trace perspective
- 14. XCF
- 15. Parallel Sysplex - XCF ā Most traffic in support of LOCK1 ā IXCLOmmm ā Telling XES Contention from False ā Latter usually caused by Lock structure being too small ā DXRnnnnn ā Telling IRLM Contention from XES ā Understand how traffic varies with time of day ā Note: XCF member job name in R742MJOB
- 16. Stored Procedures ā Native run in DBM1 address space ā Others run in Server Address Spaces ā Stored procedure defined with Application Environment ā Caller runs in a Service Class ā Queue serves combination of SC and AE ā WLM starts and stops address spaces ā PGM=DSN9XWLM ā Can see e.g. CPU, virtual storage, EXCPs, Unix statistics in SMF 30 ā Data set in SMF 42-6 ā Normally I roll up all address spaces with same name ā Probably not a good idea for server address spaces ā Not rolling up yields start / stop timestamps, balance ā Population might be important and roll up is fine for this
- 18. Applications
- 19. Application ā SMF 30 Usage Data Section
- 20. Application Performance - Using SMF 101 ā CPU usage and lots of elapsed time buckets ā With Trace Classes 1,2,3 ā Down to program (Package) with 7,8 ā SQL counts at program level with 10 ā For Interactive ā Granularity down to eg CICS Transaction and Program ā For DDF ā The SMF way to detect DDF access to a DB2 ā Lots of identifiers ā Suitable for encoding in WLM Policy
- 21. Accounting Trace (SMF 101) - Batch
- 23. Workload Manager ā SMF 30 identifies Workload, Service Class, Report Class ā I use this to establish if eg DBM1 well classified ā DDF requires Accounting Trace ā QWACWLME field ā Allows analysis of what DDF work comes in and how it's classified ā WLM ISPF TLIB tells me the rules
- 24. Rules Goal
- 25. DDF Rules
- 26. Restarts ā SMF 30 has Reader Start Time ā Can compare to IPL ā No customer (yet) restarts Production DB2 daily ā Some restart weekly or fortnightly ā Some have been up for months ā My parallel sysplex customers roll round the sysplex Gap because I have data from Weds onwards
- 27. Conclusion ā āA bridge, not a bypassā ā In a ācloudierā world detecting e.g. DB2 subsystems without specialist instrumention is valuable. ā Understanding what's normal is valuable ā Especially if you can do it with little effort
Editor's Notes
- #9: This graph plots SMF 30 data for a different customer ā for one of its Production DB2 subsystems. This is for a DBM1 address space running DB2 Version 10 and with the fix for APAR PM30468 applied (which moves the CPU from MSTR to DBM1 for Prefetch and Deferred Write engines).The three numbers are: Percent of the combined TCB and zIIP CPU that is zIIP. Percent of a processor that is zIIP. Percent of a processor that is GCP. Data is for 24 hours summarised by 15-minute interval. In this case GCPs are full speed: Both R723NFFS and SMF30SNF are 256. There is some variation in the level of zIIP eligibility ā but it is generally 70 ā 80 % of the DBM1 CPU. But there are times when it's much lower, presumably because the level of Prefetch / Deferred Write is much lower. Note: SMF30CPT does not include zIIP-eligible CPU running on a zIIP. It does include zIIP-eligible work running on a GCP. There was very little in this case.