Gil, I don't know if you have the option, but have you recently tried capping the LPAR instead, and not running a CF LPAR there? Perhaps this test would confirm or disprove your suspicions.
How does your IN/READY distribution of Queue Lengths look compared to before you started using CF Capping? Is it higher or lower? Is the job mix the same, roughly, at 100% busy times? Are you seeing a short engine effect? Does your total Logical Dispatch time show a value that's larger than the number of physical CPUs you have could possibly produce? I.E. You have 5 PCPs, you run a RMF CPU report with DINTV(0060) to get an hour summary, your CEC is 100% busy, but your Logical Dispatch total time shows something larger than 5 hours (this is possible because severely shorted engines or an overload of work to do causes a high percentage of involuntary interrupts to MVS, and the time is incorrectly accounted for by MVS)? I'm not sure what you could do the CEC to relieve your performance issues, since you seem to require 40% of the box to remain reserved, but perhaps you can look at the reports and verify that you are indeed out of gas, and perhaps compare your current results to older ones to see if you can determine if the CF is possibly to blame. Note : If you're running IRD, have you checked to make sure it's playing nicely? IRD doesn't seem to handle 100% busy situations with the grace one would expect. AFAIK, switching a PCPU from problem program mode to CF mode isn't cheap - something similar to switching from problem program mode to I/O for TPF. Best of luck, Gary Diehl ----- Behalf Of Gil Peleg Sent: Friday, December 30, 2005 10:46 AM Subject: Re: LPAR Capping There was a similar thread about this issue a couple of months ago. Since then I have been using a CF LPAR to guard 40% of the machine's MSUs by defining it to share all our CPs, and hard-cap the CF LPAR at weight 40 out of a 100 total. Lately I am seeing a very large difference between the LPAR BUSY TIME PERC and the MVS BUSY TIME PERC fields in the RMF CPU Activity report. The difference is over 20% at peak hours. This was not the case before I started using "CF capping". The CF LPAR used to guard MSUs has the highest weight on the machine. Taking under consideration the polling nature of CF LPARs, keeping the CPs busy all the time, I am starting to think this has performance implications I have overlooked before. Does this make sense? Thanks, Gil. ---------------------------------------------------------------------- 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
