Which Network architecture is correct for NOVA? and for deployment tool

Asked by Hugo Kou on 2011-03-25

Hello Guys....

http://i217.photobucket.com/albums/cc280/tonytkdk/topology.png

I draw a diagram which include 4 kinds of network architecture.

And I want to deploy multiple nodes manually or using deployment tool.

I am still not sure which kind of architecture is correct if I want to separate Nova-network & Nova-scheduler from All-in-one BOX.

Could someone help me to see this diagram , and give me some suggestions?

Any advice is good for me , plz feel free to correct my problem.

Question information

Language:
English Edit question
Status:
Solved
For:
nova-deployment-tool Edit question
Assignee:
Shigetoshi Yokoyama Edit question
Solved by:
Shigetoshi Yokoyama
Solved:
2011-03-28
Last query:
2011-03-28
Last reply:
2011-03-28
Shigetoshi Yokoyama (yoko) said : #1

I have seen the diagram. It seems that A is OK. As to B, C and D, I am
not sure about the meaning, so please give me more information.

And if you can tell me the purpose of creating such network topologies,
it will be helpful for me to consider which one is better. -- yoko

Hugo Kou (tonytkdk) said : #2

Hello Shigetoshi~

Our current corporate internal network is 172.16.2.0/24

I hope Nova to use the range 172.16.2.100-172.16.2.200 as Instance's IP. So that I can SSH / ping / WWW from other computer to Instances.

Reason why that I want to separate nova-network component!
1. I guess all instances's gateway should be nova-network server , that means all traffic will go in/out from nova-network server,
   So I try to separate nova-network & nova-scheduler into a single box. Avoid the busy traffic cause some problem for nova-api & RabbitMQ & database.

2.If I can isolate nova-network into one box , I'll try to add one more nova-network to do HA.

3.In the past , on Eucalyptus I faced some problem that while Cluster controller crash then I lost all instances. I'm trying to figure it out. So I turn to OPENSTACK.

If there has any knowing incorrect , I hope someone could correct my misunderstanding.
======================================================

My purpose is to run up instances and provide for everyone in my office.
And run WWW/APP/DATABASE server in private cloud . To do auto-scaling , live migration . for our internal work system.

I'm not sure if my consideration is correct.
There seems no any physical topology example for multiple servers on OPENSTACK.org. I try to make it.

After I success deploy one Cloudcontroller(nova-services all-in-one) with multiple compute-node , I worry about that if Cloud controller crash , I'll lose all instance. But never success to separate nova-network manually.

I found the example in DeploymentTool is what I want. So I'm going to install by DeploymentTool. I hope I can find some clue how nova-network working with DeploymentTool.

But I don't know the example's topology , I afraid that I fall into another chaos, If I use wrong topology with DeploymentTool.

Thanks

Hugo Kuo
@ American Megatrends Inc. ,Taiwan-NTG

<email address hidden>
<email address hidden>

Best Shigetoshi Yokoyama (yoko) said : #3

Regarding to the network architectures, I will comment one by one as the
followings.

A:
I confirmed it here. It is no problem.

B:
I am not clear on the difference of B and A. If the difference is just
the place connecting to external network, it is no problem in
installation. But access from external network to instances will go
through nova-api, the network communication will be concentrated on
nova-api.

C:
All the nova components are in the same segments, It should be no
problem in installation.

D:
nova-api and others are in the different segment from other nova
components. Because messages from message queue maybe don't reach other
components, I think the installation will fail.

-- yoko

Hugo Kou (tonytkdk) said : #4

Thanks Shigetoshi Yokoyama, that solved my question.