Considerations for goal selection
The following general WLM considerations can help you to properly define the correct goal definitions for Developer for z/OS®:
- You should base goals on what can actually be achieved, not what you want to happen. If you set goals higher than necessary, WLM moves resources from lower importance work to higher importance work which might not actually need the resources.
- Limit the amount of work assigned to the SYSTEM and SYSSTC service classes, because these classes have a higher dispatching priority than any WLM managed class. Use these classes for work that is of high importance but uses little CPU.
- Work that falls through the classification rules ends up in the SYSOTHER class, which has a discretionary goal. A discretionary goal tells WLM to just do the best it can when the system has spare resources.
When using response time goals:
- There must be a steady arrival rate of tasks (at least 10 tasks in 20 minutes) for WLM to properly manage a response time goal.
- Use average response time goals only for well controlled workloads, because a single long transaction has a big impact on the average response time and can make WLM overreact.
When using velocity goals:
- You usually cannot achieve a velocity goal greater than 90% for various reasons. For example, all the SYSTEM and SYSSTC address spaces have a higher dispatching priority than any velocity-type goal.
- WLM uses a minimum number of (using and delay) samples on which to base its velocity goal decisions. So the less work running in a service class, the longer it will take to collect the required number of samples and adjust the dispatching policy.
- Reevaluate velocity goals when you change your hardware. In particular, moving to fewer, faster processors requires changes to velocity goals.