multi-node setup service distribution

While I can't specifically answer all your questions, and for that I apologize, this may help:
http://nova.openstack.org/nova.concepts.html#concept-system-architecture

For Q1. Network controller be always deployed along with the Compute nodes?
I don't believe this is necessary since the messaging server is configured with a line like:
–rabbit_host=$CC_ADDR # This is where the rabbit AMQP messaging service is hosted

I got that from the multiple-server instructions which are here:
http://nova.openstack.org/adminguide/multi.node.install.html

The formatting needs some work on that page, sorry about that. :)

For Q2. What servers do you add after 2 nodes?
I'm not sure what additions would give you the most benefit. Possibly the database server? Perhaps you want to monitor the RabbitMQ traffic to see if adding another node gives the system what it needs?

http://nova.openstack.org/devref/rabbit.html has more information about the messaging.

Thanks for asking, and I'll update docs based on the information once we can answer these questions better.

2010/12/8 Shiny Sebastian <email address hidden>:
> I am trying to build a multi-node cloud using open stack. Can you tell me how to distribute the services across the nodes.

Basically, you should be able to distribute them any way you please.
You can run everything on one box or use a server per service. It's up
to you. As long as everything gets hooked up to the same rabbitmq
server and data store (MySQL or PostgreSQL or whatever you feel like),
you should be fine.

> Say for instance, should the network controller be always deployed along with the Compute nodes?

Nope, no need. You can, though.

> If not then how do we specify the location of the network controller?

You don't. They communicate over AMQP through rabbitmq. They need not
explicitly know about each other.

> I am planning to get a multi-node set up running for two nodes initially, in which case, one would be the cloud controller and the other would be the compute node as per your docs. If this works, I would want to scale it out to N number of nodes. Could you suggest some service distribution guidelines for both the cases?

I think the best we got is what Anne already pointed you to. If you
want to help improve it, that would be very much appreciated.

--
Soren Hansen
Ubuntu Developer    http://www.ubuntu.com/
OpenStack Developer http://www.openstack.org/

>> Say for instance, should the network controller be always deployed along with the Compute nodes?

>Nope, no need. You can, though.

 1.if we install the network controller on multi Compute Nodes, Who schedule the multi net controllers?
 2.In the http://nova.openstack.org/adminguide/multi.node.install.html , I am confuse with the Cloud Controller and the Compute Node ,what the real different between them?
 3.I also found that the nova-api start up from all the nodes, why it shoud be started on Compute Node ?and what the difference between the nova-api and API Server(direct connect to Cloud Controller) from http://nova.openstack.org/nova.concepts.html 's Architecture.
 4.I am confuse with those norms such as compute controller ,compute worker,compute node ,can anyone explain clear of those?