Your reply prompted me to review my notes on when and why I removed discretionary. I actually had discretionary in from 11/01 through 04. In 2004 we were literally over the edge with regards to capacity. There was quite a bit of crying about how little, or none, resources disc. was getting at that time, let alone the general sluggishness this particular LPAR was experiencing. I pulled disc. out and eventually, after flummoxing a 1 period imp. 5 mod., made it a 2 period imp. 4 to imp. 5, as I had WLM managed initiators and had trouble getting work started with the 1 period imp. 5 mod.
I should say that after reviewing my notes and some comments that Don Deese had made on this subject I would be more inclined to retry disc. Helps to go back and reread some of this stuff over time. Thanks. On Wed, 11 Oct 2006 10:08:23 -0400, [EMAIL PROTECTED] wrote: >I think that's the rub. How much can you afford in discretionary, if any. >I played with disc. early on and found that since I different requirements >for production batch workloads based on time zone and importance, and >there was very little if any cycles left for test, I had to end up putting >the test batch into a low importance service class where it would at least >get some service over time. I see this often but still just don't understand it. Say you have all this low important work with IMP=5 and some low velocity or percentile goal and you move it all to discretionary. The overall load on the system is the same and this workload should get serviced. This didn't work as well in the past as it does now, but changes over the years to how WLM manages discretionary work should allow it. Also, discretionary has the added benefit of MTTW. So I still try to put as much work as I can in discretionary. The biggest problem I had with discretionary work were the changes made to the WLM controlled initiator algorithms in z/OS 1.4 (obviously not a consideration if using JES2 inits). Since then there have been some PTFs to help this. What I did at the time (and still do) was to take that batch service class and add a 1st period with IMP=5 and a medium duration to help get the INITs started (WLM only looks at 1st period for starting INITs). This had a side benefit of getting the quick jobs out of the system if they could end while in 1st period. I used to do this in COMPAT mode for a couple of batch service classes but stopped this practice with WLM because of the (good) recommendations to try and keep around 25-30 or less "in use" periods on an LPAR. Regards, Mark -- Mark Zelden Sr. Software and Systems Architect - z/OS Team Lead Zurich North America / Farmers Insurance Group mailto: [EMAIL PROTECTED] z/OS and OS390 expert at http://searchDataCenter.com/ateExperts/ Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html ---------------------------------------------------------------------- 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
