A while back when I worked for a vendor, a large customer purchased a 10 way
because their 4 way was out of gas and they wanted the 10 way in case of
emergency during the Christmas season. We first brought one additional
engine online, then another etc and found that ALL OF THEM WENT TO 100%
IMMEDIATELY. While this was a special application, it turned out that all
engines were running the same application which acquired a lock, then
executed a long series of instructions before releasing it. Unfortunately,
since all other transactiojns were running basically the same transaction,
ALL PROCESSORS WERE SPINNING WAITING FOR THE RELEASE OF THE LOCK. When one
released the lock, the next one got it and everything went to 100 % again. I
am not saying this is what is happening, but it's a situation you don't
always see. To determine the condition, check the Excessive spin records in
logrec and check the locks held. You can also see it at the console.
Bill
From: Rick Fochtman <[EMAIL PROTECTED]>
Reply-To: IBM Mainframe Discussion List <IBM-MAIN@BAMA.UA.EDU>
To: IBM-MAIN@BAMA.UA.EDU
Subject: Re: CPU Contention
Date: Sat, 6 Jan 2007 11:18:29 -0600
------------------------<snip>-----------------------
Source : RMF Report
Earlier CPU Contention was around 99%.(when CPU Usage was 100%). Post CPU
Upgradation CPU Contention reduced to around 20%. However CPU Usage is
still hardlt 50% tht time ... what all factors contribute for CPU
Contention ?
Ideally speaking if CPU Usage is not 100% then there should not be CPU
Contention.
---------------------<unsnip>------------------------
On what part of that report do you base the conclusion that you HAVE CPU
contention?
Also, keep in mind that if multiple tasks become "In,Ready" at the same
instant, you may well have a contention situation, even if overall usage is
less than 100%. This might easily happen if an I/O operation is fielded by
interrupt, then other I/O ops are fielded via the TPI instruction. The
basic fact that CPU contention may occur isn't necessarily bad; what's more
important is how long a task has to wait for CPU service, and how often.
Both of these are much harder to measure than the fact that 5 tasks might
be waiting for service on any of 4 "engines" within the processor. And the
very nature of the workload can have a HUGE effect on all these issues. Not
to mention your WLM settings.
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
_________________________________________________________________
From photos to predictions, The MSN Entertainment Guide to Golden Globes has
it all. http://tv.msn.com/tv/globes2007/?icid=nctagline1
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html