Ask Your Question
1

Openstack+KVM+overcommit, VM prioritization

asked 2017-01-11 15:41:25 -0500

elenhil gravatar image

Hello, guys!

Imagine we have a compute node with kvm hypervisor installed

It has 64gb RAM and quad core processor

We create 2 machines in nova on this host - both with 64gb and 4 VCPUs

if both vms start using all 64gb memory, both of them start using swap

same for cpu - they use it equally

So, the question is - is it possible to prioritize 1st vm above 2nd? so the second one will fail before the 1st, to leave maximum possible perfomance to the most importan one?

like production and secondary services running on the same node.

edit retag flag offensive close merge delete

Comments

Perhaps the JSON filter can be used for that.

Bernd Bausch gravatar imageBernd Bausch ( 2017-01-11 18:39:37 -0500 )edit

as far as i understand, that's not my case I need to control perfomance priorities of already created (or during the creation) VMs, and this is a scheduler which allows me to pick some specific host to boot my VM on

elenhil gravatar imageelenhil ( 2017-01-12 01:18:28 -0500 )edit

Then I misunderstood your requirement. As far as I know, Nova or any other OpenStack module doesn't interfere with running instances, so that you have to write your own program to control this. It shouldn't be too hard.

Bernd Bausch gravatar imageBernd Bausch ( 2017-01-12 02:27:43 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
0

answered 2017-01-12 10:57:18 -0500

volenbovsky gravatar image

updated 2017-01-12 11:02:00 -0500

Hi,

for CPU - Nova resource quota should be the answer, see https://wiki.openstack.org/wiki/Insta... I also think that for CPU you can look into the direction of CPU pinning. Making one flavor (->one VM) to run over different set of physical CPU/with different ration of virtual-to-physical CPU you might achieve what you are aiming for.

-for memory corresponding functionality was not implemented. As specified in Wiki above, that is about 'memtune' of Libvirt which is not implemented in Nova as of today, https://blueprints.launchpad.net/nova...

Theoretically you can try to achieve something with having both huge pages and 4K pages on same Compute Host(s) and then you allocate Huge Pages to those 'more demanding' VMs, while normal pages to 'less demanding'.

But having both 4K and huge pages on same Compute Host has some drawback(s) and there was discussion in one of bugs that such mix is not something that design of Nova takes into account.

edit flag offensive delete link more

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower

Stats

Asked: 2017-01-11 15:41:25 -0500

Seen: 507 times

Last updated: Jan 12 '17