Ask Your Question
0

Why does OpenStack provider network setup require two compute nodes?

asked 2016-05-16 04:29:34 -0600

Muff gravatar image

updated 2016-05-16 04:30:28 -0600

Hi.

As stated in the title, why does OpenStack provider network setup require two compute nodes? Here: http://docs.openstack.org/liberty/networking-guide/scenario_provider_ovs.html (http://docs.openstack.org/liberty/net...) it is stated "At least two compute nodes" but I cannot understand why?

Why is not one compute node enough?

/Muff

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
1

answered 2016-05-16 05:48:20 -0600

dbaxps gravatar image

updated 2016-05-16 10:30:26 -0600

UPDATE MSK 17:50 05/16/2016

Relaying on already existing physical network infrastructure. They involve provider network (vlan) for creating networks connecting  Compute Nodes and Controller/Network Node. Thus this networks has IP address and MASK what requires  Router/Switch blocks like on pictures at mentioned link.Even though, if East/West traffic is not your major concern. One Compute Node might be enough.

The core target is :-

Provider networks lack the concept of fixed and floating IP addresses because they only handle layer-2 connectivity for instances.

So,they avoid traditional for Neutron L3 routing
Snapshots bellow from mentioned link is a fair for suggested design

image description

image description

I might be missing something , but http://docs.openstack.org/liberty/net...
is silently skipping the problem with cloud VMs requests to Nova Metadata Server.

I  am talking about "isolated_metadata=True" in dhcp_agent.ini.

My original answer is irrelevant to question. It presumes Neutron L3 routing in both cases. However, all fg-xxxx (Link 1) from fip-namespaces and qg-xxxx outgoing interfaces from qrouter-namespaces (Link 2) would be attached to br-int. Un-bridged external routing would come into play due to External network providers involvement.
END UPDATE

I believe that 2 Compute Nodes in case of DVR cluster are needed to verify that East-West traffic will avoid routing via Network or Controller/Network Node , otherwise see sample bellow, which might be tested on Two Node DVR cluster with External network flat provider (Controller/Network && Compute).
Running DVR with Two external networks via flat network provider on CentOS 7.2 RDO Liberty
https://www.linux.com/blog/running-dv...
Another sample with External vlan network provider, which might be tested on Two Node Cluster Controller/Network&&Compute.
RDO Mitaka & Several External Networks VLAN Provider Setup
https://www.linux.com/blog/rdo-mitaka...

edit flag offensive delete link more

Comments

Thanks for your answer. My major concern is not East/West traffic but rather North/South. So if I understand it correctly, for North/South traffic one compute node is enough?

Muff gravatar imageMuff ( 2016-05-17 04:24:49 -0600 )edit

Take a look at picture (1) 1xController/Network + 1xCompute.

dbaxps gravatar imagedbaxps ( 2016-05-17 05:34:49 -0600 )edit

You are correct. I was insufficient in my statement, I meant to include the cloud controller, but I never wrote it down. Thanks a lot for the answer.

Muff gravatar imageMuff ( 2016-05-17 05:55:34 -0600 )edit

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: 2016-05-16 04:29:34 -0600

Seen: 186 times

Last updated: May 16 '16