Ask Your Question
1

How to setup multiple Neutron servers managed by a single controller?

asked 2014-07-15 08:22:30 -0500

etlv gravatar image

updated 2014-07-15 11:03:42 -0500

smaffulli gravatar image

All multi-node architecture information I could find was talking about: single machine for controller, single machine for neutron, single\multiple machine for Nova.

There is a basic thing I cant understand. this means that if I have for example 3 compute servers running Nova on each one - all data traffic of the VMs on those 3 servers, must pass first via the same neutron server. Am I right?

This doesnt make sense from scale point of view... Can I have architecture on which I have for example, 1 controller, managing 6 Nova servers, and 2 Neutron servers (one neutron per 3 Nova servers traffic)? In other words, can I have multiple neutron servers in the same setup managed by a single controller?

edit retag flag offensive close merge delete
add a comment

2 answers

Sort by ยป oldest newest most voted
0

answered 2014-07-15 10:23:43 -0500

Rowan gravatar image

Here is the Neutron HA Architecture:

https://wiki.openstack.org/wiki/Neutron/L3_High_Availability_VRRP (https://wiki.openstack.org/wiki/Neutr...)

edit flag offensive delete link more
add a comment
0

answered 2014-07-15 11:23:24 -0500

foexle gravatar image

updated 2014-07-15 11:24:14 -0500

Which traffic you're mean ? Traffic => WAN Traffic between vm's

So traffic to WAN will pass the l3 agent where the router is mapped (l3 natting). If you have multiple l3 agents maybe on two network nodes and you have "router per tenant" option set each router will mapped on one of these l3 agents. So you can scale your traffic if you have multiple l3 agents with a router for each tenant.

Internal traffic will handle by ovs with the flow table if your instances are in the same tenant network. If you will route to an other tenant network will cross the router (routing).

On a production environment you should for every service have two servers:

  • 2 x network => neutron services (l3, lbaas, ovs, metadata ....)
  • 2 x Api (controller) => Nova-api, Nova-scheduler, neutron-server, ....
  • N x Compute => Nova-compute, neutron-agent

I hope it helps

Cheers Heiko

edit flag offensive delete link more

Comments

Hi,

Many thanks for the quick response.

Just to clarify better my question, can I have a setup which includes: 1. Dedicate server for Controller 2. Multiple compute servers - each one running nova-compute + Neutron services (l3, lbaas, ovs, metadata ....) 3. Same controller is managing all servers, and able to create via the controller VM + correlated virtual router on each server?

Itai

etlv gravatar imageetlv ( 2014-07-16 03:22:09 -0500 )edit

no thats not possible because the scheduling process of an instance will manage by nova-scheduler with your configured filter or whatever. The scheuduling process of neutron is an internal function of neutron-server. So that's not the way how Openstack works with Neutron. LBaaS, l3 agents and metadata should not present on compute nodes you should have a strictly separation between dmz and mz.

Cheers

foexle gravatar imagefoexle ( 2014-07-16 05:39:03 -0500 )edit
add a comment

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower

subscribe to rss feed

Stats

Asked: 2014-07-15 08:22:30 -0500

Seen: 1,295 times

Last updated: Jul 15 '14

Related questions

Openstack uses all available RAM

Hibernate issue on Guest VMs over Devstack [closed]

Manually created image console never appears

Instance stuck booting from Hard Disk

lbaas Unable to retrieve ready devices

Mystery subnet

Router Interfaces Down

Validation on flavor name

Error while executing neutron agent-list

Adding a second compute node fails