Does cpu ready increase when you spread it across processors?

[sjesse]

Hi

I have blades that have 2 cpus with 6 cores a piece, for the most part I have cpu ready times under 1%, but a few 8 vCPU servers are showing 2-3%

These are Terminal Servers in a horizon view enviornment and I'm trying to keep them under 5%, which from what I've read is when users start seeing issues. The other vms are all 2 vCPU, is it just the extra 6 vCPU being added causing the 2% increase, or is it the fact that I crossed over to the second cpu? These were 4 vCPU and I was having 100% cpu spikes on logon so they needed to be increased.

[flynmooney]

I probably wouldn't worry too much about 2-3% (10% is where you should start looking at ready time) and you have Co-Stop at 0.  Other than the VM showing 100% spikes were there performance issues?  After adding resources do you still have any performance issues?  It sounds like you are going over the NUMA boundry of 6 cores so for performance you might have issues with would be memory based rather than CPU.

https://itnext.io/vmware-vsphere-why-checking-numa-configuration-is-so-important-9764c16a7e73

ESXTOP - Yellow Bricks

Go download this book - http://pages.rubrik.com/host-resources-deep-dive_request.html

[sjesse]

A large group logs on once a week so I'm waiting to see if the cpu spike goes away. The horizon rdsh servers if your not aware let you provide just the application virtually instead of an entire desktop. In the metrics I see the requested apps taking 8-10 minute to open, and the cpu spike was longer than that and covered that time. I'm not worried about the 2% right now, I'm more curious why those are 2-3% and the other ones are 0.1%. I'm trying to figure out if its worth it to try 6 vCPU vs 8 vCPU, the 6 would be all on one processor. This enviornment is most likely going to expand quickly from where it is so I have to keep a very close eye on performance so I can make sure we get new hardware if we need it as soon as we determine its necessary.

[flynmooney]

Are there any disk latencies or memory issues during this time?  It might be worth grabbing the disk and memory outputs for esxtop at the same time to make sure you aren't running into something else. 

[sjesse]

Plenty of memory and disk latency was 5ms at the most.

[depping · Accepted solution]

Well 5ms of disk latency could certainly impact CPU as well. However, when you have 2-3 %RDY I would not be worried, for the number of VMs and vCPUs you are running that is relatively low. Note that when you have 8 vCPUs you have to divide 3% by 8, and then you have the per vCPU %RDY, which is extremely low in your case.

My article explains this, and also this comment on the article: http://www.yellow-bricks.com/2010/01/05/esxtop-valuesthresholds/#comment-5861

[sjesse]

Thanks, I'll review that, I haven't looked at the page since you updated it for 6.5 it seems. I had an issue a bit back where everyone was complaining about slowness and I cut back the cpu counts for everyone(cut the vcpu count in half), and everything seemed to be improved, before every vm has at least 2% up to 8%. The majority was 4 vCPU desktops with some 2vCPU desktops and 8 vCPU RDSH servers as well. I set a standard to make sure all desktops now have 2vCPUs unless they need more, which no one has.The rds servers are harder because they only need the extra cpus for when multple users logon at the same time, so I'm trying to only add what is absolutely necessary.I'm just trying to find a baseline, which seems I should reevaluate based on what you referenced. I'll review your article and I'm trying to make it through the 6.5 host resources book Frank Dennemen wrote and see if I can understand this better.