WSC IBM Z
|Abstract: This document reviews the new MCL Bundle which changes Hiperdispatch behavior for vertical CP management and how vertical high and medium CPs are assigned.|
|A new MCL Bundle on the IBM z13 (z13) processor, bundle 24, has changed how PR/SM assigns CPs in a HIPERDISPATCH=YES environment. Prior to this new bundle, any partition with a weight equivalent of between 1.5 and 2.0 processors would get at least one vertical high CP, and one vertical medium CP. Any weight of less than 1.5 processors would result in two vertical medium CPs being assigned. |
The primary benefit for this change is PR/SM is more likely to group these 2 VM processors on the same chip, resulting in better cache reuse and performance
With the new MCL on the z13, now any partition with a weight equivalent of between 1.50 and 2.0 processors will still get two vertical medium CPs. The benefit is PR/SM is more likely to group processors of the same type on the same chip, resulting in better cache reuse and performance for the processor overall.
With the new MCL in place, a CPU Activity Report will have the following layout. Notice the total logical processor share for this partition is 1.546 CPs (154.6). Before the MCL, the partition would have one vertical high CP with a share of 100%, and one vertical medium CP with a share of 54.6%. With the new MCL the share is split evenly between the two logical CPs needed to fulfill this weight.
Any partition with a weight equivalent of 2.01 CPs or higher will continue to follow the traditional CP assignment for vertical CP management with Hiperdispatch.
This new function for assigning vertical high and medium CPs is unique to the z13 and z13s model of processors. The new z14, as well as older processors will still assign a vertical high and a vertical medium for LPAR weight equivalents between 1.5 and 2.0 logical CPs.
IBM System z Family
HiperDispatch, LPAR Performance, PRSM, Vertical CP, Vertical High
|Is this your first visit to Techdocs (the Technical Sales Library)?