I had an interesting conversation today with a friend of mine who was looking for some advice about performance issues.  They had a specific virtual machine that they were running a heavy load for performance testing and were not getting the results they were looking for.  The virtual machine in question was a four CPU Windows 2003 Enterprise version with five gigs of RAM. This immediately raised red flags with me, not because the virtual machine had four CPU’s, but rather if they have one virtual machine configured that way how many more did they have in the cluster?  As it turns out they had around eight of them so I mentioned that I thought they might be having a scheduling problem and introduced them to the %ready setting from esxtop.  If you have not heard of %ready before, it is a measurement of how long the virtual machine has to wait to get access to the CPU to run commands.   When you have a four processor virtual machine, the CPU scheduler must be able to send commands to four CPU’s at once or it will wait until it can.  The higher the number, the worse the performance.  Always start off with one CPU if possible or two if you have to and work your way up from there.

Another issue they were having was getting consistent results when monitoring the host and really putting a load on some of the virtual machines.  They could not get consistent results and were very frustrated with that fact.  To make a long story short, they had DRS enabled and on a more aggressive setting. For each test, different virtual machines were on different hosts and DRS would migrate the loaded virtual machines during the process.  For the next test, I mentioned that they should set DRS to manual for the duration of any testing that needed to be done.  This will keep things the same so you can get a true understanding of the effects on the host.

ESX, Performance, virtualization, VMTN, vmware

This entry was posted on Thursday, February 12th, 2009 at 3:48 pm and is filed under Virtual Tech. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.