sjesse
Leadership
Leadership

CPU WAIT Time on Host best practice

I've never heard of this, we had someone come in to do a vCenter health check, and they checked to see if cpu wait time was under 1,000,000,000 ms. I can't seem to find anything in regards to this, we usually keep an eye on costop and ready time, but this is different.

8 Replies
sjesse
Leadership
Leadership

I mean 1 millon, not 1 billon, too many zeros

0 Kudos
dbalcaraz
Expert
Expert

what?

There is missing something, I suppose you mean a VM not the Host.

Which is the problem?

About CPU Wait Time there are resources for best practices: https://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/techpaper/performance/vsphere-esxi...

-------------------------------------------------------- "I greet each challenge with expectation"
0 Kudos
sjesse
Leadership
Leadership

No he did the host, which is all of the vms added up on one host, said they he looks for it below 1 million ms we were at like 2 million. I'm not having issues which is why I thought this was odd. I've read the perf guid before, and I don't remember this being in there.

0 Kudos
sk84
Expert
Expert

Either you misunderstood something or the guy had no idea about virtualization.

Yes, there is a value for CPU WAIT also on hypervisor level. This is the time the hypervisor has to wait for a vmkernel resource. However, this value says nothing for ESXi hosts, because it also includes the IDLE time. So it is quite normal that the CPU WAIT time is relatively high on ESXi hosts.

And apart from the fact that the value is normally given in % and not in ms, 1 million ms would be 1000 seconds. So if the ESXi kernel has to wait 1000 seconds for a resource, then you can switch off the host immediately. 😉

--- Regards, Sebastian VCP6.5-DCV // VCP7-CMA // vSAN 2017 Specialist Please mark this answer as 'helpful' or 'correct' if you think your question has been answered correctly.
sjesse
Leadership
Leadership

Thanks, I'm more leaning to "the guy had no idea about virtualization". The one we looked at is in ms though

pastedImage_1.png

and the scale isn't too far off, which is what prompted me to see what people here thought

pastedImage_2.png

0 Kudos
sk84
Expert
Expert

That's actually interesting. I wasn't aware that in the performance view in vSphere Client the CPU wait time is specified in ms. With esxtop and for the most other operating systems I know this value is given in %.

Just for fun, I have checked this value for one of our host which is currently the most busy. It has new Intel Xeon Platinum 8180 CPUs installed and currently shows about 50% CPU usage (about 70 GHz in use). It runs 70 VMs with 206 configured vCPUs. And in the performance view the CPU wait is displayed with about 4,000,000 ms. But I can't detect a performance problem with this host. Eeven with the sensitive database servers running on it.

Here are two more sources about the CPU Wait time on ESXi hosts:

Why is %WAIT so high in esxtop?

https://www.virten.net/vmware/esxtop/

--- Regards, Sebastian VCP6.5-DCV // VCP7-CMA // vSAN 2017 Specialist Please mark this answer as 'helpful' or 'correct' if you think your question has been answered correctly.
0 Kudos
dbalcaraz
Expert
Expert

Wow, it seems that someone can't be wrong here Smiley Happy

I suppose I read about VMs and this is what I put a link regarding VMs but well, I hope that sk84's response was useful.

-------------------------------------------------------- "I greet each challenge with expectation"
0 Kudos
sjesse
Leadership
Leadership

Thanks for the replies, I think I'll stick with the metrics I go by which for the most part come from ESXTOP - Yellow Bricks   with a few custom changes

0 Kudos