Capping processing weights
The PR/SM capping function provides the capability of limiting CPU resource usage for one or more processor types for one or more LP. The relative processing weight of a processor type for an LP is its capping value for that processor type.
A capped LP running at its cap for a capped processor type does not have access to the CP resources that are not utilized by other LPs. However, CP resources that are not used by a capped LP can be used by other LPs. Equitable distribution of CP resources is maintained.
Capping values can be dynamically adjusted. The capping function can be turned on and off independently for each defined processor type in an LP, and provides the capability of specifying capping for individual LPs without a re-IPL of the LP.
Use the Change Logical Partition Controls task to change the Capped setting for the specific logical partition. Selecting the Initial Capping check box turns the capping function on. Initial Capping is set independently for each processor type in the logical partition. Click Save Running System to have this option take effect immediately for an active partition. (Save and Change changes the running system and updates the partition's profile. Save to Profiles does not change the running system; it just saves the new definition to the partition's profile.) If you do not need to change a running system, the Customize/Delete Activation Profiles task can be used to open a reset or image profile to cap processing weight for an LP. This change would take effect when the partition is activated.
With HiperDispatch disabled, an LPs relative weight for a processor type is divided by the number of shared logical cores online of that type for the LP to give the share for each logical core. The goal of the PR/SM dispatcher is to give each logical core of that processor type its share of the total relative weight. Capping is done on a logical core basis.
An LPs share of CP resources for a processor type is determined by its weight for that processor type. The combined processing weights for all active LPs for a particular processor type are considered to be 100% of the available shared CP resources for that processor type. The activating and deactivating of LPs changes the amount of CP resources available to LPs, making the percentage of CP resources requested for each active LP a relative one, and not a fixed percentage of CP resources.