I think that there is a problem with Oracle on you AIX box. That much CPU is not normal.
However: The Power architecture can do more than one instruction per cycle, so at a minimum your 6 CPUs on the AIX box amount to 7200 MIPS. Our z900 with 2 CPUs is about 480 MIPS. Assuming your z890 is much faster your talking about 750 MIPS maybe. I realize that you can't directly compare MIPS on different systems, but I think your are going to have a problem. -----Original Message----- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] Behalf Of Yu Safin Sent: Thursday, June 22, 2006 7:39 AM To: [email protected] Subject: Re: SuSE Linux Benchmark On 6/22/06, Frank Schwede, LSY <[EMAIL PROTECTED]> wrote: > My 2 cents: > > Take 3-4 IFLs and put this Oracle on it. Enough for 98% of the coming > load I think... > Take 2-3 OpteronBoxes for the Java in front of it. > > AND! Put some other project on the zBox. > > Never make the mistake and size (especially) a z-machine for the peaks > you need. You also shouldn't take the Z for ONE project. > Think of the benefit of the easy consolidation and perfect system > management you will have with the Z. > And for costs you have to compare all efforts you need to get the "bit" > from the disk to the userscreens and this for a couple of projects... > > Regards > Frank > > > > -----Original Message----- > From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of > Rob van der Heij > Sent: Donnerstag, 22. Juni 2006 08:44 > To: [email protected] > Subject: Re: SuSE Linux Benchmark > > On 6/22/06, John Summerfield <[EMAIL PROTECTED]> wrote: > > > I'd rather see a benchmark involving Oracle on the Z and the Java app > on > > the {pSeries,iSeries,Sun} and if you can find a decent Opteron box, > that > > too. > > That would make sense. > And if it has to be all on zSeries, we learned that it does help to > put the tiers in different virtual machines that can be tuned > separately. Too bad many people put so little effort into it when they > do benchmarks, and then value the outcome so much... > > Rob > > ---------------------------------------------------------------------- > For LINUX-390 subscribe / signoff / archive access instructions, > send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or > visit > http://www.marist.edu/htbin/wlvindex?LINUX-390 > > ---------------------------------------------------------------------- > For LINUX-390 subscribe / signoff / archive access instructions, > send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit > http://www.marist.edu/htbin/wlvindex?LINUX-390 > Let me answer some of the questions. The benchmark will run Java and Oracle in the same box. It we get the "green light" we will split Oracle into into its own virtual machine and the Java App into a separate virtual machine. I went back to check on my AIX system and here is what I found. 1) the java application is using 15% of the cycles. 2) the remaining cycles are being consumed by Oracle. 3) we are not using Java for stored procedures. 4) the writing of the log and transactions is heavy on I/O. They do large number of deltes (400K rows) and large amounts of additions. I can't get more IFL's for the benchmark since I am using all 4 (2 for zOS and two IFL's). If we get the green light, chances are I will get the budget for a new z9. My only objective is to complete the batch processing faster on Linux than on AIX measured by my wrist watch. CPU speed is a significant factor but I suspect is not the only factor that will determine who finishes first. ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
