Yes, I do. >virtual machines method of measuring time does anybody know of a >Linux utility for performance measurement that understands that >it is working under VM. thanks
TOP, SAR, VMSTAT, etc, that produce CPU numbers from Linux on s/390 on either VM or on LPAR with undedicated processors will have incorrect CPU times. The sampling technique used by Linux to determine CPU times is very dependant on dedicated resources. Which of course does not happen on VM. As load increases, TOP and any other Linux based accounting will become MORE incorrect. The presentation I've just given at WAVV and SHARE actually shows 10 linux servers, each thinking they were using 100% of available resource. Thus all the CPU numbers provided by Linux were off by a factor of 10 - what's an order of magnitude between friends, eh? This presentation will be given at the UK CMG May 21 for Clive and friends.... And Toronto May 10th. and.... The simple solution: (Warning, vendor advertisement follows) ESALPS (Linux Performance Suite) will take the process and Linux CPU numbers, and if/when running in a virtual machine, will prorate the Linux numbers to the VM numbers, thus able to report accurate CPU utilizations by process and for each linux server. You can see http://velocitysoftware.com/esalps.html for more information. Simply, you absolutely need the VM information integrated with the Linux information to provide an accurate performance or accounting picture. >From: "C Jordan (Clive)" <[EMAIL PROTECTED]> > >We are trying to measure performance of Linux guests under zVM >using the tools available on Linux ( top vmstat sar ) along with >monitor data from CP. In an unconstrained VM system the cpu >usage as measured by VM and Linux is about the same . If we then >start up a looping virtual machine to act as a cpu soak these >figures diverge quite a lot . Rob van der Heij touched on this >over two years ago in this forum " IMHO the BogoMips should at >least be corrected for what PR/SM or CP say your unvoluntary >wait was. I would not be surprised to find more areas in Linux >where elapsed time is assumed to be equal to CPU time. ". > >Whilst accepting that this is probably a combination of the way >the CP scheduler is handling the users and also the Linux >virtual machines method of measuring time does anybody know of a >Linux utility for performance measurement that understands that >it is working under VM. thanks > >Clive Jordan "If you can't measure it, I'm Just NOT interested!"(tm) /************************************************************/ Barton Robinson - CBW Internet: [EMAIL PROTECTED] Velocity Software, Inc Mailing Address: 196-D Castro Street P.O. Box 390640 Mountain View, CA 94041 Mountain View, CA 94039-0640 VM Performance Hotline: 650-964-8867 Fax: 650-964-9012 Web Page: WWW.VELOCITY-SOFTWARE.COM /************************************************************/ "If you can't measure it, I'm Just NOT interested!"(tm) /************************************************************/ Barton Robinson - CBW Internet: [EMAIL PROTECTED] Velocity Software, Inc Mailing Address: 196-D Castro Street P.O. Box 390640 Mountain View, CA 94041 Mountain View, CA 94039-0640 VM Performance Hotline: 650-964-8867 Fax: 650-964-9012 Web Page: WWW.VELOCITY-SOFTWARE.COM /************************************************************/
