Buffer Pool Tool™ for DB2
The Buffer Pool Tool is a new generation product that allows a performance analyst to evaluate the System and Application effect of changes to DB2 buffer pools in Simulation mode – without having to actually make the changes to the production system.
Simulations PREDICT the effect of:
- Changing Pool Sizes
- Changing Pool Thresholds
- Moving Objects into Different Pools
- Moving Objects into New Pools that do not Currently Exist
Simulation / Prediction:
This component and approach allows a much faster evaluation of tuning options than a trial and error methodology, and avoids costly mistakes with the production system. Since the Simulation shows the impact and the interaction of all objects together within a pool, is much more effective and accurate than attempting a pool isolation methodology for tuning. Quite directly, this is the only way you can get it right.
Temporary pool isolation, and performance evaluation of an object in one pool might take you half a lifetime to complete for a large DB2 environment, and the fallacy of pool isolation is the lack of interaction between multiple or many objects. This cannot be determined or evaluated using isolation techniques.
Statistical Analysis
The Statistical Analysis component provides a level of information for the overall System, each Pool, and every Object, that is simply not available from any other product. Additionally, a PC based graphics component makes it easy to determine which objects are performing the best, which are performing poorly, the impact each is having on its current pool, and which objects will obtain the greatest improvement from changes.
A PC based graphic analysis component processes summarized data from the mainframe facilities, and easily shows the best and poorest performing pools and objects. It also provides analysis of the current environment and makes tuning suggestions for your system. These suggestions may encompass anything from changing pool thresholds, increasing pool sizes, and moving objects into different pools. The software is Y2K compliant.
A Coupling Facility sizing component is available that provides initial sizing specifications for the CF structures in a Data Sharing Environment. This PC based component facilitates the easy evaluation of various sharing options, and uses the statistical data from the operating DB2 subsystem rather than the redbook guesstimates of “small”, “medium”, or “large”.
| Product Functions | BPT | Online Monitors | DB2 Pool Displays | Platinum SubSystem Analyzer |
|---|---|---|---|---|
| Predicts System Performance Levels at Varying Pool Sizes | Y | |||
| Predicts Improved Object Performance within each Pool | Y | |||
| Move objects into different Pools in Simulations | Y | |||
| Adjust VPSEQT and HPSEQT Thresholds in Simulations | Y | |||
| Object Access Information, Rate and Type of Access | Y | Y# | Y# | |
| Object I/O Rates, Avg. & Max. Elapsed for all I/Os | Y | Y# | Y# | |
| System I/O rates, Avg. & Max. Elapsed for all I/Os | Y | Y# | Y# | |
| Processes a Meaningful Data Collection Interval > 30 Minutes | Y | Y* | ||
| Extensive Statistical Analysis at Pool Level | Y | |||
| Ranked Statistical Analysis for each Object within each Pool | Y | |||
| Statistical Object Usage Analysis for Plan, Authid, or Combination | Y | |||
| Provide Average and Maximum Resident Pages by Object | Y | Y@ | ||
| Average Page Residency Times | Y | |||
| Helps Identify Transient Objects | Y | |||
| Shows how to reduce Transaction Elapsed Times | Y | |||
| Optimize DB2 Memory Usage, Eliminate I/Os, and Save $$$ | Y |
* The data must be available for every object in every pool during the same time interval to be meaningful. It’s usually impossible to display all of the data using the Buffer Pool Display commands for pools with hundreds of objects. Since the displays do not order the output by access rates, one or more highly accessed objects may be lost because the line output is limited.
If multiple detail level displays are utilized, they’re not from the same time interval and aren’t statistically relevant.
# Usually cannot handle a long enough trace period to provide statistical relevance
@ Can show the maximum (not an average), and can’t indicate when this happened; within the last five minutes, five hours, or five days