vSAN CPU sizing

[RolandoRodrigue]

Hello, dear community

Please help me with this:

I was sizing a vSAN cluster with workloads residing on a cluster of a very old version of processors (E5 2630 v3). I understand that 1 core of, for example, Xeon Silver 4208, could be better in performance that the old processor I mentioned. So, there is a way to calculate how much better (numerically way) are the new processors?

I was doing some research and I found that Nutanix applies some "weight" to the physical cores. So depending on your new processor, let's take 4208 8C, you can have (3-node cluster dual processor) 48C + applied weight. So, at the end you will have "more physical cores" to cover the vCPU required by the workload.

I noticed that with the vmware sizer it generates over sized configurations, so I was wonder if there is other way to size vCPU on vSAN.

Thanks in advance!

[TheBobkin]

Hello Rolando,

" So, there is a way to calculate how much better (numerically way) are the new processors?"

There are various CPU benchmark comparison sites available online e.g.:https://www.cpubenchmark.net/compare/Intel-Xeon-E5-2630-v3-vs-Intel-Xeon-Silver-4208/2386vs3507

Intel Xeon E5-2630 v3 @ 2.40GHz vs Intel Xeon Silver 4208 @ 2.10GHz [cpubenchmark.net] by PassMark S...

"So, at the end you will have "more physical cores" to cover the vCPU required by the workload."

vSphere works more with guidelines on what is considered optimal vCPU to physical core ratios - the general rule-of-thumb is that one should generally not over-subscribe more than 3-4:1 vCPU:core but this is not a hard rule that should be blanket applied to all workloads e.g. some latency-sensitive and/or compute-intensive workloads may not perform optimally unless they have 1:1 ratio (and adequate number of vCPUs assigned, though more vCPUs doesn't always increase performance, it can in some cases do the opposite).

"I noticed that with the vmware sizer it generates over sized configurations, so I was wonder if there is other way to size vCPU on vSAN."

These being 'over-sized' may be based on conservative assumptions of what are typical optimal amounts for the workloads types specified (or perhaps you are under-estimating VMs CPU resources!), or this could be as a result of what the typical storage layout per node is in conjunction dual-socket set-ups meaning there is a fair bit extra CPU (in which case, consider lower cost/performance/core-count CPUs).

Sizing workloads for ESXi vs vSAN is no different other than having to account for ~10% overhead for vSAN:

Designing and Sizing vSAN Hosts

Bob

[RolandoRodrigue]

Hello, TheBobkin​

Great tool for CPU comparisson! It will be very helpful.

Yeah, I knew the sizing is the same as well on ESXi plus the vSAN overhead. I was missing the fact that new generation processors gives more performance than the older ones. So, If the CPU mark shown in the tool is almost 1.5x higher for the Silver 4208, it means that 1core of Silver 4208 could be equivalent to 1.5 core of E5-2630v3? And using that assumption to cpu overcommit, it means that an overcommit of 4:1 (on E5 2630v3) coud be equivalent in performance to a 6:1 on Silver 4208?

Thanks!