number of cores vs virtual sockets on 6.0 and 6.5

[mdangel1]

All,

We have always increased the number of virtual sockets instead of cores per socket. That has been the recommendation. I came across the following article which states to do the apposite.

Has VMware changed their recommendations in 6.5? That is what it seems like.

https://blogs.vmware.com/performance/2017/03/virtual-machine-vcpu-and-vnuma-rightsizing-rules-of-thu...

[daphnissov · Accepted solution]

The link to the comment you posted does not support the statement that cores should be preferred over sockets. It's still recommended that you configure sockets over cores unless you have a licensing-specific need.

[mdangel1]

Thank you for your reply. This is what is confusing me: please the underlined:

In all the documentation and VMware books I have read, our rule of thumb has always been change the number of virtual sockets and not the number of cores per socket

I propose the following Rules of Thumb:

1. While there are many advanced vNUMA settings, only in rare cases do they need to be changed from defaults.
3. Always configure the virtual machine vCPU count to be reflected as Cores per Socket, until you exceed the physical core count of a single physical NUMA node OR until you exceed the total memory available on a single physical NUMA node.
5. When you need to configure more vCPUs than there are physical cores in the NUMA node, OR if you assign more memory than a NUMA node contains, evenly divide the vCPU count across the minimum number of NUMA nodes.
7. Don’t assign an odd number of vCPUs when the size of your virtual machine, measured by vCPU count or configured memory, exceeds a physical NUMA node.
9. Don’t enable vCPU Hot Add unless you’re okay with vNUMA being disabled.
11. Don’t create a VM larger than the total number of physical cores of your host.

[daphnissov]

Those rules of thumb are for vNUMA considerations. And if you look at the graphic of the table Mark provides further down the article, it shows only cores per socket when the vCPU count passes 10 based on the NUMA layout of this particular CPU and model. Otherwise, and only until that point, you should configure sockets over cores.

[IT_pilot]

I agree with daphnissov

I studied the behavior of NUMA on a dual-processor 12-core host (vSphere 6.0).

Described my observations in the article, but it is in Russian.

The idea is that within the 1 Node of NUMA you need to increase the sockets.

And then (24 CPU and more in my case) a single-core configuration can show worse performance than a multi-core configuration. I attach a screenshot of the table for information.

In version 6.5, NUMA works much better.