Our environment meets all three of the items on your bulleted list. Our main production environment is VM/CMS (with DB2, ESAWEB, huge SFS farms, etc.), and that workload is on general purpose CPs, not IFLs. LINUX is the only workload on the IFLs, and we (presently) have only one production server so we WANT it to exploit ALL available resources (there is actually a second one, for TEST, but its access to IFL resources is controled via traditional SET SHARE).
-Mike -----Original Message----- From: The IBM z/VM Operating System [mailto:ib...@listserv.uark.edu] On Behalf Of Rob van der Heij Sent: Thursday, August 20, 2009 7:52 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: z/VM mode LPARs On Thu, Aug 20, 2009 at 1:27 PM, Michael Coffin<michaelcof...@mccci.com> wrote: > PS: Yes, we've given each Linux server 5 virtual CPU's, which map to > two real IFLs. Our thinking on this is that the Linux guest "should" > dispatch more work simultaneously with 5 CPU's (and the two real IFLs > can more than handle the load). Comments? Good idea/bad idea? Pretty bad idea. Increases cost and makes performance worse. You never need more virtual CPUs than you have real ones. http://nl.wikipedia.org/wiki/Stoelendans But even that number is for most systems too much and useless as rule of thumb. Your Linux server needs one virtual CPU to run. You add another one only if all following are true: - your workload can use multiple CPUs (not every workload can) - business justifies that it uses that amount of resources (in your case, that one guest uses more than half of your CPU resources on its own) - total utilization is low enough that there is a reasonable chance it would get more than one full engine when it wants Rob - "When you don't know, one will do. When you have measured, probably too." -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/
