Sounds like a reason to have a two-period batch service class. The short
jobs will always be counted (will never be outliers); get some idea of
the 90th percentile SUs for the short jobs and make that the duration of
the first period, with a RT goal of between two and five seconds
depending on measured experience. Second period can have a reasonable
response time goal e.g. one minute at the 80th percentile. Refine based
on experience.
For important work the goal should be achievable but not ridiculously
so. WLM should have some work to do.
I think I've said all these things before...
Steve Samson
Tom Marchant wrote:
On Wed, 19 Sep 2007 10:30:41 -0500, Dave Kopischke wrote:
... I can also
imagine several hundred very short duration JOBs skewing the total
performance of a service class and impacting WLM's management of that
class.
That's why you need to be careful in selecting the percentile. You want WLM
to exclude the outliers when it makes it's calculations.
----------------------------------------------------------------------
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
