z/OS supports the use of general purpose processors, called CPs along with two types of specialty processors called zAAPs and zIIPs. z/OS will determine if it has any of these processors available and if so, will attempt to offload a certain amount of work to them. IBM customers usually pay for their mainframe software usage by means of a monthly variable charging mechanism, similar to a cell phone bill, where the more you use it, the more it costs you. This is true only of usage of CPs. Usage of zAAPs and zIIPs, once you have bought the hardware, is unmetered, so the variable monthly charges can be lowered if eligible work is offloaded from CPs to zAAPs or zIIPs.
IBM provides a broad range of capacities within a particular mainframe model family, without the need for hardware intervention, by reducing the full rated speed of CP engines by different amounts.
A useful attribute of zAAPs and zIIPs is that no matter what the running speed of the CPs, zAAPs and zIIPs always run at the full rated speed of the processor model.
This is great for users - but - if there is a difference in processor speeds it means that you cannot directly compare the utilization of CPs with zAAPs or zIIPs. A CPU second on a CP will be “smaller” than a CPU second on a faster zAAP or zIIP. To help with this IBM provide a conversion factor in SMF type 72 records that allow you to calculate “CP equivalent” time, and the calculation is :
CP equivalent seconds = zxxP seconds * conversion factor / 256
You can see this conversion factor in the Acquire output listing in message AZS722I. If the value is 256, then your CP processors run at the same speed as your zAAP or zIIP processors. If it is larger than this, the specialty processors are running faster than the CPs. For example on a z10 model 2098-Q02, the conversion factor is 710, which means that one CPU second used on a zAAP or zIIP processor would take 710 / 256 = 2.77 CPU seconds on a CP.
Acquire converts system-level, WLM-level and job CPU zxxP data to be CP-equivalent times. Syncsort Capacity Management provides the means to display them and the un-normalized versions of this data, so you can choose what you want to see. Partition-level information is always stored as-is, and is never normalized.
This means that when looking at normalized metrics the system-level or sum of Service Class or Report Class zxxP CPU times or CPU % will be higher than that seen at the partition level. It may even display values over 100% busy. This may feel “wrong", but all it means is that your zxxP utilization, if run on CPs, would need more power than the machine currently has available.