What is the best architecture for quantum

asked 2013-04-25 16:10:20 -0600

mrCloud gravatar image

updated 2013-05-15 04:21:36 -0600

fifieldt gravatar image

Hi Folks

I am trying to design a production openstack solution. Does anybody have any suggestion of best production openstack architecture. I am looking things like, if I should have network node on a dedicated server or should I install network node along with compute node? My assumption is more modular is better and seperating duties on multiple server may help scalling further?

This is mostly for Grizzly. Can any of you share some thoughts. So far I am thinking 3 node setup, node1 controller, node2 compute and node3 network quantum. Is this a good architecture to follow? Ofcourse there are lot of components which could also be seperated?

Which one of these nodes can be run as VMS or have you run any? Thanks

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2013-05-10 03:57:55 -0600

Robert van Leeuwen gravatar image

We are running with a pretty similar setup:

  • Openstack controllers
  • Compute nodes
  • Routing/network nodes

We do not run compute along the Controller and Network nodes. You can do it but since these are such important components we do not want any chance of a virtual machine going haywire impacting those machines. Also it keeps the setup simpler making it easier to troubleshoot.

There is a very good reason to not run the quantum dhcp/l3-agent on the same machine: http://docs.openstack.org/grizzly/openstack-network/admin/content/ch_limitations.html "If the host does not support namespaces then the quantum-l3-agent and quantum-dhcp-agent should be run on different hosts. This is due to the fact that there is no isolation between the IP addresses created by the L3 agent and by the DHCP agent. By manipulating the routing the user can ensure that these networks have access to one another."

Note that the l3-agent will be a single point of failure if you setup networking to route all traffic across it. There was work being done to make it HA but it seems it did not make the Grizzly release. I'm hoping they will back-port this at some point. You can also setup Pacemaker for HA but I'm personally not a big fan of cluster software since it adds quite a lot of moving parts which tend to cause more downtime than it's worth.

If you are going for a really big setup you can run every service on a different node (keystone, glance, nova etc.) and running multiple instances of the service and use a load-balancer to spread the API calls (everything is HTTP so easy to load-balance).

edit flag offensive delete link more

Get to know Ask OpenStack

Resources for moderators

Question Tools


Asked: 2013-04-25 16:10:20 -0600

Seen: 321 times

Last updated: May 10 '13