Question about over VM consolidation number and CPU Ready percentage

[AlbertWT]

Hi People,

I'm in the process of consolidating as many VM into single ESXi servers running on HP Blades on AMD Opteron with 32 logical available CPU (based on the number on the vSphere console)

Can anyone here please share some thoughts and comments about what is the best indication to avoid adding more VM than the ESXi hosts can handle ?

Do I have to make sure that the number of vCPU in total not exceeding the 32 logical available CPU or it can be more than that ?

Running a powershell script from LucD to gather the CPU ready percentage shows that some VMs got CPU ready more than 10% (mostly 2x vCPU) up to 22% (this one with 8x vCPU)

[DavoudTeimouri · Accepted solution]

Hi,

I had same situation on HP Blade and DL serers with two different CPU vendors.

Based on my experiences, Intel CPU has better performance compare to AMD CPU when CPU is over-committed on ESXi server.

I have HP BL460c G6 and G7 on one of our sites, the server has 12 physical cores and 24 logical core.

I assigned four cores to each clients (For resolving voice issue) and the VMs are working fine, CPU RDY is under 5% typically and there is no performance issue.

Each Blade server is hosting 20 VMs and total virtual cores are 80 cores, so we assign 1:3 (Physical:Virtual) core.

On other hand, I have DL585 G7 on our some sites and the server have 64 physical cores (4 sockets and each sockets have 16 cores). The servers are hosting 50 active VMs usually but CPU RDY is bigger than 10% or 7% on them. Total virtual cores are 200 and CPU assignment ratio is 3.

I think, if you want best performance on AMD CPU, 1:1 or 1:2 CPU assignment ratio is best choice.

[homerzzz]

Are the end users experiencing any performance impact? If not you are still good. % ready for an 8 vCPU VM does not seem bad...only ~3% per vCPU. You will not find an exact ratio as an answer since it depends on the workloads and end user feedback.

For example, my current vCPU:pCPU ratio on hp gen8 blades (Intel) is 5:1 and I am not experiencing any CPU related performance impact.

[AlbertWT]

Thanks for the reply guys,

So in this case based on the following screenshot below:

The suggested total maximum amount of vCPU that can be assigned to all VMs running in the ESXi host is 1:3 or from the highlighted value above 24*3 = 192 vCPU (192 VMs with 1x vCPU) ?

is that correct ?

[AlbertWT]

yes, the end user complaints that the Windows Server VM running Oracle Forms (Java) even though I have already assigned 8x vCPU it still runs very slow.

[homerzzz]

Is this screenshot current? If so, the performance issue with your Oracle Forms server is not due to overprovisioning vCPUs.

I think your math is a little off...1:3 equals 24:72. Remember, that vCPU number will not be a hard maximum since there are many variables that will determine the maximum amount of vCPU your environment will handle.

[AlbertWT]

yes it is the current screenshot.

if that is the case, then what could be the culprit ?

[homerzzz]

Is the Oracle Forms VM a new VM or was it running somewhere else previously?

I do not have any Oracle Forms experience, but I would make sure the guest OS is properly configured and I would also review the Oracle performance tuning docs.

Are your blades configured to best practices for ESXi....power management settings etc. ? What are you using for storage?

How is the performance with the other VMs?

[AlbertWT]

Homer,

The Oracle form server is Windows Server 2008 R2 it is a new implementation since 2 months ago. The problem is that it runs very slow and lagging when users connect remotely to this server to generate report as it is running its own Java Virtual Machine.

THis is HP BL465 G7 Blades.

The storage is using SSD on PureStorage array. So it is not a problem in terms of storage.