On Friday, 07/28/2006 at 11:47 AST, "Wakser, David" <[EMAIL PROTECTED]> wrote: > You have asked the correct question, since I am pretty certain that the SIE > is the problem (older VMs had 3, and that was reduced). Any answers to the > question: will replacing a single LPAR running z/VM (and 2nd level z/VMs with > 3rd level guests) have any advantage?
There will be performance advantages, yes. There will be administrative disadvantages. This is the trade-off. vSIE (the thing invoked when the 2nd level system issues a SIE instruction to dispatch a 3rd-level guest) is slower than the real SIE instruction. Further the SIE instruction the 1st level VM system ultimately issues on behalf of that 3rd-level guest has so many provisos, advisories, and limits, that it will exit at the slightest provocation. And the time slice for a 3rd-level guest is a fraction of the 2nd level time slice. The 3rd level guest just doesn't typically get to spend much quality time alone with the CPU (relative to all that's happening in the 1st level system). You can see, I hope, why it can be difficult to maintain production performance levels for 3rd level guests. You can do it, but it's labor and resource intensive. For development, testing, and QA, 2nd level VM systems are the bee's knees, but my preference is to put production VM systems in LPARs. The z9 EC has 60 LPARs, in case you were wondering. Do you want one? I know people..... And because this is a discussion about performance, my attorney advices the following disclaimers: - It depends(sm) - YMMV :-) Alan Altmark z/VM Development IBM Endicott