Enabling HiperDispatch for z/OS Logical Partitions

HiperDipatch is enabled under z/OS by default. To disable HiperDispatch under z/OS, specify HiperDispatch=NO in the IEAOPTxx member of SYS1.PARMLIB. However, logical partitions with greater than 64 logical processors defined at IPL are forced to run with HIPERDISPATCH=YES. After IPL, partitions with greater than 64 logical processors are unable to switch into HIPERDISPATCH=NO. Also, when partitions specify LOADxx PROCVIEW CORE, they are forced to run with HIPERDISPATCH=YES. Refer to the z/OS publications for further details. There are no new hardware controls or settings to control use of HiperDispatch within a logical partition; however, WLM management of logical processors normally works best with global performance data security setting enabled. When global performance data security setting is disabled, one should usually be conservative in the number of logical processors defined in excess of the share of the type of processor (CP/zIIP) . The share of a type of processor in a partition is expressed as the number of physical processors that a partition is entitled to by the weight of the partition. For example a machine with 8 general purpose CPs and two partitions with weights of 600 and 400. The share for partition with weight 600 is 600 / (600 + 400) * 8 = 4.8 physical processors. If one defines 8 logical processors for this partition, 5 will be active at all times and up to 3 will be active only in periods that WLM determines 1 or more of the 3 has a good opportunity to increase the net capacity consumed for this partition. In the example, if the second partition is using its share of 3.2 physical processors, having one or more of the 3 excess logical processors in the first partition is not sensible in most cases since there is no excess abandoned share from the second partition.

For partitions with global performance data active, WLM is able to manage the excess logical processors that are active to intervals where other partitions have not completely used their share. When global performance data is disabled, WLM uses all logical processors at all times which may lead to attempts to consume processor that can not be achieved. The efficiency of the partition may be less with 8 logical processors active at all times in the example above. Defining 6 logical processors for the first partition is normally a better choice. If for some period of time the 8 logical processors are likely to use abandoned share from other partitions, 8 is a good number.