On Tue, 23 May 2006 00:00:00 GMT, Ted MacNEIL <[EMAIL PROTECTED]> wrote:
>>The concern is trying to adjust WLM for the different hardware, i.e. a velocity of 30 on the 2094 means something different than velocity 30 on the 2064 within the same plex, correct ? > >In theory, no. Quite the opposite. And it's more than just theory, it's fact. Have you ever read this white paper from John Awre? MVS Workload Manager Velocity Goals: What You Don't Know Can Hurt You http://www-03.ibm.com/servers/eserver/zseries/zos/wlm/documents/velocity/velocity.html "Third, the observed velocities of multitasking address spaces can be sensitive to the number of CPUs in use by MVS. At equal CPU utilizations, systems with different numbers of CPUs running heavy multitaskers like DB2 will have very different CPU queue lengths. Probability theory states that with a CPU utilization of 90%, at any given instant on a machine with one CPU there is a 90% chance that all CPUs are busy (and hence a 90% chance that newly ready work will be queued). At the same utilization on an otherwise identical machine with five CPUs, the chance that all CPUs are busy is (.90) sup 5 or 59% if the state of each CPU is assumed to be independent. The table below shows some observed values running the DB2 Large System Performance Evaluation (LSPR) workload on the same processor family with the DB2 address spaces 3 running in service class SYSSTC." Table 1. N-way Effect on DB2 Velocity +----------------------------+---------+---------+ | Processor Model 9021 | 711 | 952 | | Number of CPUs | 1 | 5 | | DB2 Achieved Velocity | 13.3 | 74.9 | | SYSSTC Achieved Velocity | 17.4 | 74.7 | | CPU Busy % | 89.75 | 88.60 | +----------------------------+---------+---------+ >The only differences would be CPU service, and that's supposed to be normalised by using service units. > >I have mixed 9672's and z/900's in the same SYSPLEX with no ill effects (a long time ago). > Ted's experiences here are still valid. I have been in shops with this type of mixture, but it can be a challenge for workloads that are set up to use velocity goals in a mixed environment. Sometimes I've had to go against the "single image" concept of sysplex and set up different goals by lpar where I could. Mark -- Mark Zelden Sr. Software and Systems Architect - z/OS Team Lead Zurich North America / Farmers Insurance Group mailto: [EMAIL PROTECTED] Systems Programming expert at http://expertanswercenter.com/ 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
