Uncaptured CPU

Uncaptured CPU is the difference between the "Total LPAR Effective Dispatch CPU Time" and the "Total CPU Time Captured" at the LPAR level.

In MXG and SAS language terms using the RMFINTRV member it can be calculated from the following variables:

UNCAPT = CPUEFFTM - CPUTM ;

There is also a capture ratio variable in RMFINTRV (named CAPTURAT) which should be tracked in concert with Uncaptured CPU . Conversely you can derive Uncaptured CPU using the following calculation:-

UNCAPT = CPUEFFTM * (1 - (CAPTURAT/100));

Uncaptured CPU time can then be converted to MIPS via these variables

UNCPMIPS = (UNCAPT *(capacity/#cpus))/DURATM ;

where

capacity = rated MIPS capacity of the processor (we use Gartner MIPS ratings)
#cpus    = number of physical online general purpose CPUs in the processor (excludes special purpose engines, such as ICFs & ZAAPs)

The use of MIPS is very important when changing processors as this normalises the quantity of Uncaptured CPU based on the processor's attributes. For example, when moving from a z900 to a z9 EC the CPU time actually reduces due to the faster CPUs in the z9 EC, so using MIPS (or MSUs if that is your site standard/choice) will give you a consistent comparison before and after processor changes, and any other change for that matter.

NB: If there is a difference in the above calculations then this may be due to LPAR management overhead (from LPARNAME='PHYSICAL' segments in TYPE70PR ) attributed in the calculation using the CAPTURAT variable. I separate this overhead out by using : LMT = CPUACTTM-CPUEFFTM;

As some or all of the new special purpose engines (ICF, IFA, IFL, ZAAP, ZIIP, ...) have their own 'PHYSICAL' segments, you may need to exclude them using SMF70CIN="CP " for the general traditional CP. This is still an evolving area which complicates things so refer to the latest IBM sources as well as MXG newsletters/changes/release levels to determine what you need in place to support data from z9 EC processor and other generation models. There are a lot of changes and considerations in this area to be across for reporting purposes, dependant on your site's configuration and needs. It is best to actually look at the data in TYPE70PR and modify code and reporting accordingly; eg. I had to change code for differing values of LCPUADDR for z9 EC and z990 models.

I find it much better to work in MIPS for Uncaptured than percentages for the capture ratio since for the latter if you do not know the magnitude of the total then percentages do not give the full picture, whereas the actual MIPS value is easy to compare.