kvm cpu usage allocation

kvm cpu usage allocation

[Alejandro Comisario]

Hi everyone.
We are having a private cloud with more than 1000 phy servers.
Each server has 128GB of RAM and 24 cores (2 heaxacores with HT
enabled), but the amount of vms running on those servers versus the
flavor (amount of CPU) made us use about 80 VCPUS.

The questions are as follow.

* is there a fine documentation of HOW kvm handles cpu allocation/pin
per vm in an scenario like this one ?
* is there also, a documentation that explains under a circumstance
where a VM needs more processing power, how priority is managed ?
* is there a way to define some kind of politics or rules to have
specific cpu pinning for specific vms ?

So, well ... any extensive information to understand how KVM behaves
would be great!
Thanks as allways!

----------
@lejandrito

Re: kvm cpu usage allocation

[Alejandro Comisario]

any help on this one ?
thanks!


@lejandrito

On Wed, Feb 19, 2014 at 6:42 PM, Alejandro Comisario
<alejandro.comisario@mercadolibre.com> wrote:
> Hi everyone.
> We are having a private cloud with more than 1000 phy servers.
> Each server has 128GB of RAM and 24 cores (2 heaxacores with HT
> enabled), but the amount of vms running on those servers versus the
> flavor (amount of CPU) made us use about 80 VCPUS.
>
> The questions are as follow.
>
> * is there a fine documentation of HOW kvm handles cpu allocation/pin
> per vm in an scenario like this one ?
> * is there also, a documentation that explains under a circumstance
> where a VM needs more processing power, how priority is managed ?
> * is there a way to define some kind of politics or rules to have
> specific cpu pinning for specific vms ?
>
> So, well ... any extensive information to understand how KVM behaves
> would be great!
> Thanks as allways!
>
> ----------
> @lejandrito

Re: kvm cpu usage allocation

[Marcelo Tosatti]

On Wed, Feb 19, 2014 at 06:42:56PM -0300, Alejandro Comisario wrote:
> Hi everyone.
> We are having a private cloud with more than 1000 phy servers.
> Each server has 128GB of RAM and 24 cores (2 heaxacores with HT
> enabled), but the amount of vms running on those servers versus the
> flavor (amount of CPU) made us use about 80 VCPUS.
> 
> The questions are as follow.
> 
> * is there a fine documentation of HOW kvm handles cpu allocation/pin
> per vm in an scenario like this one ?
> * is there also, a documentation that explains under a circumstance
> where a VM needs more processing power, how priority is managed ?
> * is there a way to define some kind of politics or rules to have
> specific cpu pinning for specific vms ?

Hi Alejandro,

Check "Chapter 7. Overcommitting with KVM" of


https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Virtualization_Administration_Guide/chap-Virtualization-Tips_and_tricks-Overcommitting_with_KVM.html