scalability of compute server more than 2000

asked 2017-09-28 19:47:26 -0500

ejbaek gravatar image


I'am searching useful idea to handle more than 2000 compute servers by one openstack controller node.

I am trying to seperate messaging queue and DB from openstack controller node to separated server to cover 2000 hosts. And then I am testing performance. But it looks dividing messaging queue and DB is less effective. Because even though seperate messaing queue and DB, load is high on openstack controller node.(cpu, network traffic, memory, etc.)

If you have idea to handle 2000 hosts using one openstack controller, please leave comment behind.

edit retag flag offensive close merge delete


There used to be a large scale deployment group; perhaps they still exist. Reach out to them:

While I have no experience with deployments of this size, I think a single controller is totally unrealistic.

Bernd Bausch gravatar imageBernd Bausch ( 2017-09-29 22:30:51 -0500 )edit

Check out the HP Helion "mid-scale" model for a couple of hundred nodes, far away from your 2000. Metering, DB and message queue, Network controllers are each separate clusters:

Bernd Bausch gravatar imageBernd Bausch ( 2017-09-29 22:33:05 -0500 )edit