Workload manager LPAR CPU management of shared CPs

WLM's LPAR CPU Management component, together with the LPAR clustering technology of the z17, provides the ability to dynamically manage workloads within an LPAR cluster comprised of multiple logical z/OS® images on a single z17. Each LP is assigned a transaction goal (desired response time) and an importance level. WLM monitors how well each LP is achieving its goals. A donor/receiver approach is utilized to reapportion CPU resources between LPs in the cluster. When WLM LPAR Weight CPU Management decides to change the weight of an LP, it adjusts the receiver LP and the donor LP by a percentage of the current weight of the receiver. WLM takes resources away from an LP that is over-achieving its target or has a workload that is less important (as defined by the installation). Any resource given to a particular LP is taken away from another LP in the LPAR cluster. LPs whose workloads are of the same importance level should all have similar performance indexes (a measure of how closely the workload is meeting its defined goal).

One can think of the entire LPAR cluster as having a total processing weight. The total weight for an LPAR cluster is the sum of all the initial processing weights of all the LPs that have joined the cluster. As a new logical partition joins the cluster, its initial processing weight is added to that of the cluster. Though weights are adjusted from one LP to another, the total weight for the cluster is consistent. When an LP leaves the LPAR cluster, as when it is either system reset, deactivated, or re-IPLed, the initial processing weight, which it had been contributing to the LPAR cluster, is removed from the total weight available to the cluster. The weight removed from the cluster is not necessarily equal to the current weight for the exiting LP.

The optional minimum and maximum processing weights for an LP govern how much flexibility WLM has in adjusting weights from one LP in the cluster to another. The installation should assign a reasonably wide range of processing weights to each WLM managed LP. Assigning the same value for initial, minimum, and maximum weights effectively disables WLM LPAR CPU Management of processor weights.

Though logical cores of WLM managed LPs may need to be soft-capped (as for workload pricing, see Workload charging by soft-capping to a defined capacity), initial capping (traditional hardware capping) of these LPs is disallowed. Similarly, if an LP is WLM managed, its logical cores must be shared, not dedicated. For more information regarding internals of WLM CPU Management, see the IBM Redbook z/OS Intelligent Resource Director