P2V and possible CPU contention

[Sperwer70]

Hi all,

I am struggeling with CPU performance.

I have 2 ESX servers that i want to use for MS SQL servers. I want virtualize our remaining current physical SQL servers. Memory isn't an issue, but there are some concerns about possible CPU contention.

From vSphere i can deduce the total amount of used cpu resources in Mhz by the current virtual SQL servers, but i need to get the same sort of info for the physical servers.

Now, through the windows performance monitor on the physical machines i can get my (_Total)\%Processor Time, but how does this relate to cpu usage in Mhz as shown in vSphere.

Is there a way to convert the windows info to Mhz?

Martin

[MKguy · Accepted solution]

Is there a way to convert the windows info to Mhz?

Well, you can simply calculate the effective Mhz used if you know the physical host's total CPU capacity:

([total number of physical cores] * [core Mhz]) / [% total CPU usage]

This should give you a rough outline of the required CPU usage.

It should be noted that different types of CPUs can have some effects as well:

For example, if your physical server is running old Nehalem generation CPUs and your ESXi hosts run newer Ivy-Bridge Xeons, the same workload probably requires less CPU cycles on the ESXi host with newer CPUs because of new/improved instruction sets, more efficient pipelines, larger Cache sizes, faster memory bus etc etc.

The same obviously applies vice-versa if your ESXi host runs older CPUs than your physical server.

[Sperwer70]

Hi MKguy,

Thanks for the answer, this help a lot.

Luckily the physical servers are quite old (one of the reasons for virtualizing them) and the ESX servers are recent.

Martin