vCPU scheduling

[timbo475]

I've read a bit around this subject but would benefit from a bit of help from people who really know this stuff because I'm a bit confused.

I was on an ESX3.x course and the instructor mentioned something about it not being a good idea to over resource VMs with vCPUs due to scheduling gotachas. Reading around the subject the strict co-scheduling that ESX .2x used would definitely have created problems with some VMs using multiple vCPUs under certain conditions but I've read that the co-scheduler is relaxed and this means that this problem is a bit less apparent.

We are running an ESX 3.0 server with 3 VMs and 4 vCPUs. One VM is has 1 vCPU assigned, 1 has 2 and the other has 4. I suspect the VM with 4vCPUs is having major performance issues (effectively, it's wanting to use all available vCPUs which is probably bound to cause scheduling issues) but I'm unsure how to troubleshoot CPU scheduling issues.

How strict is cosheduling in ESX 3 and what's the best way to troubleshoot vCPU scheduling issues?

Cheers,

Tim

[weinstein5]

With the current release of ESX 3.5 the vCPU scheduling has improve in that the vmkernel will schedule each vCPU when it has space available - with ESX 3.0 the scheduling is similar to esx 2.x - the vmkernel will not schedule any vCPU of a multi vCPU VM if it can not schedul all vCPUs - so if you have a vm with 4 vCPU running on an ESX 3.0 host with 4 logical CPUs - you will have performance problems with that VM - look at the real time performance graphs and check what the CPU Ready measurement are - this is a measure of time the VM waits for the vmkernel to schedule the vCPUs - so if there is a problem schedulling you will have a high ready time -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful