Ask Your Question

Scaling limit of nova computes connected to a single neutron controller?

asked 2014-01-27 21:17:47 -0600

stack555 gravatar image

Hi All,

How many nova compute nodes can a single neutron controller handle?


edit retag flag offensive close merge delete


Good question. Atleast I have not seen this kind of figure yet.

dheeru gravatar imagedheeru ( 2014-01-28 00:04:44 -0600 )edit

I wanted to know it either. How much routers, floating IP's, network throughput it can handle.

Drankis gravatar imageDrankis ( 2014-07-14 04:29:15 -0600 )edit

3 answers

Sort by ยป oldest newest most voted

answered 2014-07-15 08:47:50 -0600

gmi gravatar image

If you watch these two HP presentations you'll see that they talk about running 3-4000 virtual routers in one networking node, so the limits can be high if you tune the configuration accordingly:

Of course, your results might be different because of a number of factors, so you should always test your assumptions and have a plan for upgrade when you reach the limits.

edit flag offensive delete link more

answered 2014-07-14 10:44:11 -0600

mpetason gravatar image

updated 2014-07-14 10:44:33 -0600

There are limitations based on the OS with how many open files/sockets. If you are asking how many can your Network node handle before you need to stand up multiple services then you would have to add in a bunch of variables to get your answer:

What speed is your network? Are you bonding network interfaces and using 10G?
What are the specs of your hardware for the network node?
Are you running your neutron API on the network node too, or just l3/dhcp/*aas?

You're going to get different answers from different people since everyone has different configurations.

edit flag offensive delete link more


Ok, If I got fast network, fast interfaces, single l3 + dhcp service on ubuntu machine can it handle, for example, about 300 routers and about 2000 floating ip's without crashes?

I understand, that there are a lot of variables, but some main limits should be...

Drankis gravatar imageDrankis ( 2014-07-14 11:19:43 -0600 )edit

If it has been configured correctly then you are looking at throughput and then OS hard limits. If you are hitting the limit on throughput then you would want to setup another network node to act as an endpoint for specific networks.

You'd probably still run into crashes with 30 routers and 200 floating IP addresses. You could tune as you go, or what you really want to do is look into HA neutron (DVR) :

I haven't seen any hard set numbers. I've seen configurations that have between 1-3 with thousands of VMs and they don't run into networking related issues due to not enough hardware.

mpetason gravatar imagempetason ( 2014-07-14 11:46:18 -0600 )edit

answered 2014-07-14 04:33:48 -0600

SGPJ gravatar image

updated 2014-07-14 04:42:11 -0600

In devstack; the limit is 10 instances and 50 floating IPs but not sure on Openstack.

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



Asked: 2014-01-27 21:17:47 -0600

Seen: 674 times

Last updated: Jul 14 '14