Distribution of high load within clusters, single thread limited to physical CPU (& memory) ?

[AiwarikiaR]

I have not been able to discern from the documentation I have read so far how the load from one VM is distributed should the load exceed the bounds of a single CPU core.

At what level is load distributed between physical CPU cores and more importantly between physical servers?

At the level of Virtual Machines,

At the level of processes,

At the level of threads, or

Somehow at a sub-thread level?

I will ask in another way; does the ESX provide virtual CPUs that surpass the capacity of single CPU cores or single servers, that also has the ability to run one single thread that consumes the entire capacity of the virtual CPU?

Can someone point me in the right direction?

Is there some good documentation for this issue?

-AiwarikiaR

[mcowger]

No - a virtual CPU given to the gues has EXACTLY (unless you artifically limit it) the same abilities as a single physical core.

So a given VM with (say) a single vCPU will run exactly as fast at the physical core it is executing on. It will not cross core/sock/machine boundries to make a larger, faster virtual CPU.

On other words, for your second workding, the answer is 'no'.

--Matt

[AiwarikiaR]

So if I am running a server installation which has a heavy load on say one (1) HP DL380, there would be no benefit beyond failsafe/redundancy if I created an ESX resource pool containing two (2) HP DL380 servers (i.e. no performance increase) ?

And can I ask again for pointers to somewhere this may be written down in documentation, a FAQ or otherwise?

Thanks for the reply mcowger.

[mcowger · Accepted solution]

I dont think thats untrue, unless you have only 1 VM.

If you have more than 1VM, having for CPUs to service them will almost always help performance.

best place for doco like this is the VMWorld transcripts and the VMware classes.

--Matt

[weinstein5]

A virtual CPU runs on a single core at a time - so once a virtual cpu is using all of a physical core that is all it will get -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

[AiwarikiaR]

I just have to squeeze the last drops out of the lemon:

Since virtual machines can be created with at least two virtual CPUs, is it possible to create virtual machines that utilize physical cores in more than one physical server?

And what is actually the limit for # of virtual CPUs in ESX? (I have only used VMware workstation previously)

I suspect the conclusion will be that server clusters and resource pooling is good for many "smaller" virtual machines, while boosting performance for single, load heavy VMs will require investment in high performance hardware. And there might be little benefit from running such monsters as VMs at all...

Thank you for your help.

I was holding off on the awards to elicit more responses

-AiwarikiaR

[mcowger]

Since virtual machines can be created with at least two virtual CPUs, is it possible to create virtual machines that utilize physical cores in more than one physical server?

No

And what is actually the limit for # of virtual CPUs in ESX? (I have only used VMware workstation previously)

4 per VM. 192 total vCPUs in a ESX host.

And yes, your conclusion is correct- ESX isn't designed for making 10 machines act like 1 big one, its designed to consolidate many workloads that dont deseve a whole machine into 1.

--Matt

[weinstein5]

A VM can only run on a single ESX server at a time - from the configuration maximums document on http://www.vmware.com/support/pubs/vi_pages/vi_pubs_35u2.html an ESX Server can have 192 virtual cpus - your conclusion is correct yes you can virtualize any machine but does it make sense - in the case of a heavily loaded physical machine like a SQL server processing many transaction it might not -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful