RMF Meaning of "Dispatch Time"
The PR/SM Logical Processor Data section contains the single most important measurement that is needed to analyze PR/SM LPAR system activity: dispatch time.
rm
Total logical processor dispatch time is the sum of the time intervals during which the logical processor was dispatched on a physical processor. Total dispatch time includes LPAR management time that is attributable to the specific partition. If LPAR management time reporting is in effect, effective dispatch time is also reported in this section. Effective dispatch time does not include LPAR management time. The difference between total and effective dispatch time is the LPAR overhead.
It is important to understand the meaning of the word "dispatch" in the PR/SM LPAR sense (the meaning is not the same as when it is used to describe address space activity within the system). LPAR dispatch time is accumulated when the logical processor is available for use by the partition, regardless of whether the processor is actually busy. This time includes wait time if wait-assist is enabled and a fixed-time-slice value is specified.
In other words, logical processor dispatch time is CPU busy time only if wait-assist is disabled. If it is enabled, you must subtract the value of the CPU wait time recorded in the CPU Data section from this dispatch time to arrive at CPU busy time.
Also important to understand is that there is no permanent correspondence between logical processors and physical processors except for the brief duration of a time slice, because a ready logical processor can be dispatched on any available central processor. RMF may be accumulating measurements for what it regards as CPU 1, but the actual address of the central processor assigned as CPU 1 may be any of those available at the time.
Individual central processor utilization levels in a PR/SM LPAR complex cannot be calculated from RMF data. However, by adding the dispatch times from all of the active logical processors in the complex, it is possible to calculate complex-wide utilization, or average processor busy.
Each PR/SM Logical Processor Data section also contains the logical processor address and the wait-assist status. The LP address is essential when calculating processor utilizations for wait-enabled partitions because, while the dispatch time is recorded in the logical processor section, processor wait time is recorded in the CPU Data section. The wait-assist status is available in the form of a bit within a status byte that also contains indications of a change in the wait-state status or processor relative share within the measurement interval.
The processor relative share represents a mechanism for allocating processor resources in a CPC during periods of high utilization and contention, or when partition capping is in effect. It is a dimensionless number, the absolute value of which is meaningful only when compared to the relative share value of the other active logical partitions in the complex. If a partition has dedicated processors, the recorded value of the processor relative share is 65,535 (hexadecimal FFFF).