Comments interspersed.

From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf 
Of Klein, Kenneth E
Sent: Wednesday, May 14, 2014 3:36 PM
To: [email protected]
Subject: supervisor state vs. problem state

What is the general opinion out there on what portion of CPU cycles should be 
spent in supervisor state?
Is more than 50% a "bad thing"?
---> See my previous reply to Gil. I would consider this a bad thing on a z/OS 
box. On an *IX box, not necessarily. The beans are counted differently and 
stuff that z/OS charges back to a user, *IX boxes just account for as 
"overhead" (i.e. supervisor state). The RMF workload activity report should be 
able to give you some help here.

Could having a logical processor to physical processor ratio of greater than 2 
(e.g. 8 logicals in a 3 CP cec) be "bad"?
---> PR/SM overhead grows dramatically after 3:1 logical to physical. CPU times 
increase due to cache flushing in the processor (aka context switching). Your 
comment above equates to 2.67:1
I have run for several years with a 3:1 with minimal issues IMO, the 2.67 to 1 
is "OK". YMMV.

What is a reasonable rate for SIGP (Signal processor) instructions? Our rate is 
in the thousands.
---> Don't know. The most probable answer is "it depends". IO configuration, 
processor enablement for IO interrupts.  Logical to physical processors,... all 
have a role to play

HTH,

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to