I was asked this question; I asked around and Ive summarized the responses below.
z/OSMF has two address spaces. The Angel task and the server task
The Angel task
Once the Angel task has started it uses very little CPU. It provides cross memory services for z/OSMF.
If your default service class for this is SYSSTC, then this will be fine.
The server
This is a more interesting challenge. This address space runs all of the work. It is written in Java. On my system it takes a lot of CPU to start up, but once it is up, it only uses CPU when there is work to do. This work tends to be as a result of a web browser or REST requests, and is not usually mission critical.
Running this as with a service class of Discretionary. May be OK, but as z/OSMF is used by systems programmers, they may give it a little extra boost.
If you are using z/OSMF to run and monitor your system, you may want to put it in a service class with non Discretionary, so that it continues to run even it your system is constrained for CPU.
Personally I would use TSO and ISPF in preference to z/OSMF, but new people to z/OS are familiar with Web Server based products rather than TSO.
I asked about this problem on the IBM Mainframe mailing list,and got some responses.
Kolusu Sri said
IZUSVR1 uses the setting from IZUPRMxx parmlib member, so you can add WLM_CLASSES in there.
On my system we have
WLM_CLASSES DEFAULT(IZUGHTTP)
LONG_WORK(IZUGWORK)
you also need racf permissions
SEC_GROUPS USER(IZUUSER),ADMIN(IZUADMIN),SECADMIN(IZUSECAD)
SESSION_EXPIRE(495)
The IZUGHTTP service class is for foreground z/OSMF transactions (except asynchronous beans).
The IZUGWORK transaction class is for long running asynchronous beans and for unexpected z/OSMF work, not otherwise classified. Asynchronous beans are used in some scenarios of the Software Management plugin.
See z/OSMF V2.1 Resource Requirements section WLM Classification for z/OSMF on page 11. For more information.
Mike Shorkend said
Also keep in mind that a large proportion of IZUSVR1 CPU consumption is ZIIP eligible. I prefer to send ZIIP heavy workloads to a dedicated service class (mixing workloads might cause a ZIIP eligible workload to get denied ZIIP because the service class it is associated with is meeting its goals because of other non-ZIIP workloads). On a busy development LPAR I use a service class with importance 5 and execution velocity of 30 and it performs well. Also keep in mind the HONORPRIORITY setting for the service class which can cause/prevent spill over of ZIIP eligible work to general usage processors.
ColinPaice 