On Fri, 2007-06-15 at 14:47 -0500, Al Sherkow wrote: > You're really off here... > > The premise of your question is wrong, so it's hard to answer the question.
As Jim and Al have said, things don't work the way you postulated. In fact, to use the vernacular - "you've got it arse about". When you get dispatched, you get 100% of an engine. When you a see a processor at (say) 60%, that is *NOT* the engine running at 60% - it is the processor running at 100% for 60% of the time. Extrapolate across all the (logical) engines in the LPAR. Plenty has been said in the past about the limitations of using averages based on a sampler, but it's what we have. When you see your LPAR (of say 5 engines) at 60%, it is *not* running on 60% of the engines. It is running on all 5 for 60% of the time. Effectively all 5 are running at 60%. This leads to "short engines" (do a search) - and is a likely cause of the performance impacts you are seeing. Presuming no IRD involvement - others will be able to say whether IRD is even definable for capped LPARs; I haven't looked. Whatever, don't blame WLM and the LPAR dispatcher - the cap was established as a business decision by your company. It may be that if your cap (and presumably target weight) resolve to approximately 60% of the CEC you'd be better off reducing the number of engines to ameliorate the short engine effect when capping is enforced. If the cap is just too tight for the work to get through, swallow hard and make a (business) decision. 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
