Choosing service class periods

z/OS® Capacity Provisioning uses the performance index (PI) of service class periods as the primary trigger for provisioning and deprovisioning actions. For example, the Capacity Provisioning Manager considers a provisioning action only if the actual PI of any included service class period is worse than the defined PI. You must select appropriate service class periods for which the performance index is correlated to the performance of your business application.

For example, an important business application can consist of multiple service classes that are defined with importance 2 and 3 in the WLM service definition. Somewhat simplified, the goal attainment of those service classes primarily depends on other work that is classified as importance 1, SYSSTC or SYSTEM. When system resources such as processing capacity become constrained, WLM attempts to help the most important work first and assign fewer resources to other work. In this example, resources are taken from the service class periods with an importance of 3 first, and they in turn show a higher (worse) PI.

It is best practice in WLM not to activate too many service class periods at any point in time. There must be a substantial service measurement in any period so that WLM has sufficient sample data. Having sufficient amounts of sample data is especially important for service class periods that are defined with average response times or response time with percentile goals. The same objective applies to Capacity Provisioning. The number of transaction endings should be high enough to allow your monitoring product, such as RMF™ to compute a performance index. At least one transaction ending is required for each gathering interval, such as the RMF MINTIME, and preferably more endings guarantee for significant interval data.

You can specify service class periods with importance levels of 1 - 5 only. SYSTEM, SYSSTC, SYSSTC1 to SYSSTC5 and DISCRETIONARY work cannot trigger a provisioning action and is ignored, if included. Do not include any service classes that are associated with a maximum capacity for a resource group. If a service class is capped, Capacity Provisioning avoids provisioning on behalf of it, if possible.