Although they aren't exact numbers, mainframe capacity planners find CPU hours, MIPS and MSUs good indicators for...
plotting future growth. Reducing MIPS is also a huge priority for mainframe users who want to contain software costs.
Mainframe shops can use these handy shortcuts to learn about CPU, MIPS, mainframe application capacity planning and converting CPU hours to MIPS.
On the mainframe, MIPS is generally a measure of computing capacity. IBM publishes the relative MIPS for each processor family through its Large System Performance Reference, which is based on statistical analysis of what IBM considers representative workloads. However, a mainframe system's actual performance depends on its IT workload profile and the number of installed engines. This is especially true for IBM's z13 hardware offering, which sacrifices raw clock speed for bigger caches and more sophisticated instruction pipeline processing.
Million service units (MSU) is the measure of machine capacity generally used for calculating software licensing costs. MSU ratings are closely related to MIPS, although IBM sometimes manipulates MSUs to encourage mainframe users to upgrade to new hardware. To account for the multiprogramming effect, IBM assigns MSU ratings for different processor models on a sliding scale depending on the number of active processing units -- also referred to as CPs by IBM.
To convert CPU seconds (accumulated consumption) to MIPS (average consumption speed), the capacity planner divides the equivalent uniprocessor MIPS (EMU) by the elapsed seconds, then multiplies the result by the CPU seconds. That result is the MIPS. EUM is defined in the REstructured eXtended eXecutor language's REXX exec.
CPU seconds to MIPS conversion example
A mainframe job has used 100 CPU seconds during one minute -- it is a multitask job. The system is an IBM 2064 Model 1C5 (1085 MIPS, EUM = 217 MIPS). The average consumption speed of this job is 100 x 217/60 = 362 MIPS.
Cutting CPU costs on the mainframe and improving the system's operational efficiency, however, are not always the same thing, which leaves room for debate about what metrics, including CPU hours and others, to share with the business.
Editor's note: This was first published in February 2012 and updated in February 2016.
Capacity plan for unpredictable IT workloads
Become a z13 SIMD instructions master
How mainframe demographics change today
Dig Deeper on Data center capacity planning
Related Q&A from Robert Crawford
I have two years of experience in mainframe technology, currently working as a mainframe developer. I want to change to Java technology. Continue Reading
I want to replicate DB2 from the mainframe to an AIX box since it's cheaper and the copy can be used for testing. Is this possible? Continue Reading
What's the impact on I/O when you run in a CPU-constrained mainframe environment? Continue Reading