This article looks into some of the steps necessary to locate and tune DB2 application performance problems. It will:
- Discuss the tools and structures needed for DB2 tuning.
- Address the identification of DB2 performance problems, and the initial reports and processes used for this identification.
- Explain some of the detailed investigation required to alleviate performance problems.
DB2 Tuning Information Sources
Prior to any DB2 tuning, there must be a review of the performance information that can pinpoint the problem. DB2 sub-systems have facilities to gather information and produce DB2 TRACE records which are written to either SMF or GTF destinations. The normal destination for performance-related informational records is SMF. DB2 creates three common SMF record identifiers: 100, 101 and 102. DB2 assigns multiple classes to these record types. These classes are specified in DSNZPARM or on the START TRACE command, and tell DB2 what level of detail to accumulate.
SMF 100 records are called Statistics records and contain DB2 system-wide information. Four classes are associated with statistics records. Since very minimal overhead is associated with statistics records, the Rule-of-Thumb is to turn on statistics classes 1, 3, 4 and 5. These classes contain the following information:
- Class 1 – These records are written every "N" minutes (as specified in DSNZPARM) and show summary information
- related to all DB2 activity during this time interval.
- Class 3 – These records are written for every deadlock or timeout and contain information related to these incidents.
- Class 4 – These records are written during DB2 exceptional conditions and contain diagnostic information.
- Class 5 – These records contain information related to Data Sharing.
SMF 101 records are called Accounting records and contain the individual DB2 thread-related information used to initiate any performance tuning project. The Rule-of-Thumb for accounting trace classes is to turn on 1, 2, and 3 (if you're using DB2 packages and would like package level accounting information, activate classes 7 and 8). The following explains the normal accounting classes:
- Class 1 – This record contains basic accounting information.
- Class 2 – This record contains "in-DB2" time. Without this class it's impossible to determine if the problem resides in DB2 or in another resource or product.
- Class 3 – This record contains wait time. This is important if the problem is related to buffer pools, CPU overload, DASD, or other factors that contribute to wait time.
- Class 7 – This record contains Package/DBRM basic accounting information.
- Class 8 – This record contains Package/DBRM level wait time information.
SMF 102 records are called Performance records and contain detailed performance-related information. These records are collected for very detailed problem determination. Beware, some of the performance traces seriously affect DB2 performance.
Other Performance Data
Another important source of tuning information is DB2 EXPLAIN output. This information describes the methods that DB2 uses to retrieve answer-sets. It's very important to always run EXPLAIN when binding plans or packages. It's also very important to retain historical data of EXPLAIN output. This history is vital in determining if a performance issue is related to data retrieval method changes.
The final base information that's useful for DB2 performance tuning is RMF data. The RMF information can be especially useful if the problem is CPU resource or DASD contention. RMF helps to pinpoint the specific problem for these types of constraints.
Identify and Isolate Performance Problems.
Where should you start if it appears that there are DB2 application performance issues? The first thing to do is identify that there truly is a problem. This may be recognized by:
- A user's complaint that things are taking too long.
- A performance monitor.
- A proactive (normally user written) accounting information analysis application.
Isolate the Cause
Once a problem is recognized, the 80/20 rule should be used to identify the small percentage of DB2 transactions that represent the majority of the total workload. These transactions will give the greatest payback when they're tuned. During this phase of the investigation ask questions such as: is this problem related to a single transaction or job or user, or does this problem only appear at certain times of the day? Getting the answers to questions like these will narrow the list of "offending" transactions to a manageable number.
DB2PM Short Accounting Report
During the tuning project's fact-finding phase, the normal starting point is to review the DB2PM Short Accounting Report. This report provides indications of resources used. The short accounting report will identify:
- DB2 response time (elapsed time).
- Resources used (processor and I/O).
- Lock suspensions.
- Application code changes (via the SQL used fields).
- Wait times (processor, I/O wait or lock wait).
These fields identify potential problems. There is one warning when using the Short Accounting Report: this report is the average of all executions of a plan or package for the reporting interval. Averages can have the effect of hiding some very long running executions of the plan/package if 1) there are many executions of the plan/package during this period and 2) most of the executions are of very short duration. Again, look for the plan/package(s) that appear to be using the majority of resources during the reporting interval.
Detailed Performance Investigation
Once the "problem" plan/package has been identified, it's time for a detailed review of the information related to this plan/package. The first place to look is the EXPLAIN output. Run a stand-alone EXPLAIN to determine the current access path that DB2 is using. Then compare this information to the EXPLAIN output (accumulated in your PLAN_TABLE) as it relates to the last bind that was run for this plan/package. If the access path DB2 is using is incorrect based on how you think DB2 should be retrieving information, check for things such as missing statistics in the DB2 catalog or changes in SQL calls.
DB2PM Long Accounting Report
If the DB2 access path review doesn't reveal the solution, examine the DB2PM Long Accounting Report. This report gives detailed information regarding plan/package execution. Following are a few of the key pieces of information presented in this report:
- Class 1 Elapse Time – Compare this value to CICS or IMS transit times. They should be close, but not necessarily identical since DB2 time doesn't include:
- Time before the first SQL statement.
- DB2 create thread time.
- Time after the DB2 terminate thread.
- Not-in-DB2 Time – This is the calculated difference between Class 1 and Class 2 elapse time. If time spent outside DB2 (but within the DB2 accounting interval) is lengthy, the problem will be found in the application, CICS, IMS, or the overall system, and not within DB2.
- Lock/Latch Suspension Time – This value shows contention for DB2 resources. Check the "Locking Summary" section of this report for additional information, then proceed to the Locking Reports for help.
- Synchronous I/O Suspension Time – This is total application wait time for DB2 synchronous I/Os. If the number of I/Os is high, check for:
- A change in access path.
- Application code changes.
- System-wide DB2 bufferpool problems.
- RID pool failures.
- System-wide EDM pool problems.
- Asynchronous Read Suspensions – This is the accumulated time for read I/O done under a thread other than this thread. It includes time for Sequential prefetch, List prefetch, Sequential detection or Synchronous read performed by another thread. The Rule-of-Thumb for Sequential prefetch or Sequential detection (asynchronous I/O) is 1 to 2 milliseconds per page. The Rule-of-Thumb for List prefetch is 3-4 milliseconds per page. Check "Other Read I/O" to locate value.
- Not-Accounted-For DB2 Time – This is accounting class 2 time that is not part of class 2 CPU or class 3 suspensions, and is normally due to MVS paging, processor wait time or time spent waiting for parallel tasks to complete. Check the "Not Account" field for this value.
There are many places to look and many causes for DB2 performance-related issues. This article has been a good starting point for tuning DB2. You will find that the better you understand how DB2 functions, the easier performance tuning becomes. For additional information on DB2 application tuning, go to the following information sources:
The opinions in this article are solely those of the author, and the information herein is to be taken "as-is".
This was first published in November 2005