VM cpuunits with multiple cores

[jared555]

How are cpuunits balanced when virtual machines have different numbers of cores?

For a simplified example, say I have a two core hypervisor. There are three virtual machines running on the hypervisor. One has two cores and two have one core. All three have 1024 cpuunits, unlimited limit.

In a maximum load scenario, will each machine receive 2/3 of a core or will the two core machine receive an entire core and the other two machines half a core each?

I am trying to figure out a billing scenario where I allow a user to create multiple machines with a maximum combined total cpu power. Ex: Planning for 1024 cpu units per hypervisor core, allow a user to create one machine that uses 2048 units or four that get 512.

 

[wolfgang]

Hi,

the problem is this are relative and not absolute values.
What means if you have 3 VM's with A 512, B 1024 and C 2048 units.

C have double so much cpu time as B and four times more than A.
But this only works if all your cores are on 100% usage.
This means if your cpu are bored all VM get 100% of the core what they have.

If you need more information search for
"cgroups cpu.shares"

 

[jared555]

Thanks for the response. I just wanted to make sure cpu shares were allocated on a per VM basis and not a per thread/core basis. Seems that per thread cgroup control has been removed so it has to be per VM.