> We have as 2064-2C5 with 3 LPARS weighted as follows: > > SYSA - prod 415 uncapped 5 cpus online > SYSB - dev 75 capped 5 cpus online > SYSC - sysprog 10 capped 2 cpus online
I don't like capping - goes against the grain. Similarly, with a low target for SYSB, you are guaranteeing "short engines" there. On average, that LPAR will only be getting 3% of each of the 5 engines. If you were 2 trim it to 2 engines, you'd get 7.5% of each. I'd bet you'd find the work throughput would increase. Easy enough to bring them back online if you cause problems. > We have a single TCB CICS region running on SYSA that "maxes out" at about > 83% (415 / 5 engines) at most peak times. (CICS sometimes also goes max > tasks due to CPU constraint when SYSB is fully loaded. We then change the > weight for SYSA to over the 415 for a certain amount of time to let it > catch up.) > > Our SYSB is suffering lately for CPU in the above configuration. At > times, SYSA is not maxed out and we would like to give SYSB more CPU > dynamically. We are worried, however, that by raising the weight or > uncapping SYSB we may cause the high CPU CICS region to suffer on SYSA. > If, for instance, we took off the cap on SYSB and it went to a weight of > 100, then SYSA would get only a 390 weight. The high CPU CICS region > would possibly suffer because of its single TCB architecture. True? > Thoughts? Set your weights to reflect what you want to achieve if all LPARs were to go 100% at the same time. You seem to misunderstand how the weights work. If SYSB were to go over its target, and then SYSA hit a demand load, all LPARs would be driven back to the weight proportion. SYSB would go back to 15%, and SYSA would ramp up. Let the LPAR sheduler do its job - if you get the numbers wrong, adjust them till you're happy. Shane ... ---------------------------------------------------------------------- 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
