Hard capping and the impact on z/VSE
Ingolf24 120000DRN3 Visits (4195)
Today it's just a short discussion on LPAR (logical partition) capping and how it may impact z/VSE.
A flavor of virtualization is integrated into System z servers provided by the Processor Resource/System Manager (PR/SM). PR/SM allows to divide the servers into logical partitions (LPARs). Each LPAR get assigned a portion of available physical resources, which can be shared with other LPARs or dedicated to a specific LPAR. You may create up to 60 LPARs running z/OS, z/VM, Linux, TPF or z/VSE. z/VSE may run in a LPAR or as z/VM guest on z/VM in a LPAR.
Capping can be used to limit the CPU consumption of an LPAR. This may be necessary, if you want to ensure that your LPAR (and z/VSE system) only gets a specific amount of MSUs (Million Service Units) - or MIPS.
Reasons can be that you want to
There are for sure more reasons.
Each partition has a relative weight (set via HMC - Hardware Management Console), which affects the available resources to a LPAR. You may enforce a relative weight, that is PR/SM never allows to use more shared resources (CPUs) than specified in the relative weight (hard capping or intial capping).That is you may limit the MSUs available for the LPAR workload.
If z/VSE runs in a hard capped LPAR, you may need to adjust a few system parameters. Dependent on the available CPU resources
z/VSE partition balancing - especially balancing of online and batch workload - should be considered. You may adjust the PRTY (priority command) shares or lower the MSECS command value, which defines the interval in milliseconds after which the balanced group is inspected as long as the time slice is not exhausted. By default the value is about 1 second. You may set it to 100 (milliseconds).
Dependent on your workload there may be more tuning efforts necessary - or you may need to increase the relative weight.
If you use the QUERY TD command or the IUI system activity display to monitor the CPU consumption of your workload, you will see the CPU utilization that corresponds to the relative value - and not 100%, if your workload uses all hard capped CPU resources.