On Tue, 18 Sep 2007 14:17:59 -0500, Kelman, Tom wrote: > >I have also used response time goals for batch work, but only for batch >that should be short running (short test jobs) and where there would be >a relatively large number of jobs in a given time period. Then I have a >second period to drop the work into and give that a velocity goal if it >goes past the desired turnaround time. In doing have used that to >penalize programmers who submit jobs to my "short job" (30 minutes or >less) job class that will end up running for hours. My thoughts on >needing a large number of jobs going through are that if you don't have >enough work then WLM won't get enough samples to calculate a valid >response time. Am I thinking correctly here? >
And I have some online work that I've assigned velocity goals. I think assigning a goal has more to do with the nature of the application. This particular CICS region has transactions that perform millions of I/O's. That doesn't fit into the response time goal model at all. Identifying the kinds of work you have is going to help define your service classes. Knowing what's important is going to allow you to classify your work. This is the role of an SLA. You can still guess at what's important and move forward without a formal SLA. Although I'm not an expert, I believe WLM captures performance data several times every second. Even very short duration JOBs will be seen. But 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. You really need to know a little bit about the nature of the work and classify it accordingly. ---------------------------------------------------------------------- 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
