Quantum with vSwitch on 2 Compute Nodes

Hi Sunil,

A lot of your questions might be answered by checking out Quantum documentation at http://docs.openstack.org/incubation/
Documenation specific to Open vSwitch Plugin for OpenStack Quantum can be found here - http://openvswitch.org/openstack/documentation/

It is a little difficult to get to these docs as Quantum is a 'core project' for the upcoming Folsom release and in incubation for current Essex release.

I'll try to briefly address some of your questions here as well( all these are based on the assumption that you are running OpenVSwitch plugin for Quantum):

1) Flags for second compute node:

--libvirt_ovs_bridge=br-int
--libvirt_type=<libvirt type, e.g. qemu>
--libvirt_vif_type=ethernet
--libvirt_vif_driver=nova.virt.libvirt.vif.LibvirtOpenVswitchDriver

2) Floating IP:

Floating IP is supported by Quantum as of Essex-3 release. So Floating IP will work with Quantum just like it works with Nova networks. The only change is Floating IP will be attached to the "Virtual networks" exposed by Quantum instead of being attached to traditional nova networks. You will still attach Floating IP to a nova network but these nova networks will be "virtual networks" powered by Quantum.

3) In the default implementation of Open vSwitch Plugin, each Quantum network is just a VLAN on your network, and that's how the traffic will flow from Open vSwitch to another Open vSwitch.

The documentation links provided above should answer any other details regarding Quantum set-up and Administration.

Thanks for the reply. I am new to Open Stack and treat me as an end user for now.

I have read the docs and would read them again but have some conceptual questions as the docs are very confusing. ;-)

I like to see diagrams. ;-)

If we can speak, it would be better.

(1)

I am running devstack stack.sh to install s/w.

For running stack.sh on second node, what flags I need? I am not sure of what all components I need to install when running stack.sh.

I could narrow down to these but that is still a larger combination.

ENABLED_SERVICES="n-api,n-cpu,n-net,n-vnc,openstackx,quantum,q-agt"

(3)

Assume you have two Compute Nodes connected on 2 Ports of an Upstream Access Switch.

We create a network L2 N1, and launch Guest 1 on Host 1 and Guest 2 on Host 2.

So will the two Guests G1 and G2 be on same L2 Network N1, or we have to create a seperate L2 Network N2 for G2.

How will L2 connectivity work between G1 and G2 in regards to switch ports? How does Physical Switch know VLAN Tags for N1 to allow Trunk mode? Or they will be internal to Hosts and not span Vswitches?

How will L3 connectivity work in regards to NAT and routing for VMs? Think the Nova Network on Node 1 would be acting as Router/Default Gatway for Node 2, and do NAT, but it brings into dual NIC topology and without switch direct connect.

I was thinking of all Compute Nodes being connected to Switch with one NIC.

How does two Vswitches talk with a Switch in middle

(2) For Floating IP

http://docs.openstack.org/diablo/openstack-compute/admin/content/associating-public-ip.html

Will this be Ok example to follow?

I was not sure if G1 and G2 could share a L3 Subnet on different nodes.

Here a topology diagram becomes very important as there are many ways to connect L2/L3.

On Thu, Mar 8, 2012 at 2:00 PM, Sunil Srivastava <
<email address hidden>> wrote:

> Question #190097 on quantum changed:
> https://answers.launchpad.net/quantum/+question/190097
>
> Status: Answered => Open
>
> Sunil Srivastava is still having a problem:
> Thanks for the reply. I am new to Open Stack and treat me as an end
> user for now.
>
> I have read the docs and would read them again but have some conceptual
> questions as the docs are very confusing. ;-)
>
> I like to see diagrams. ;-)
>
> If we can speak, it would be better.
>
> (1)
>
> I am running devstack stack.sh to install s/w.
>
> For running stack.sh on second node, what flags I need? I am not sure
> of what all components I need to install when running stack.sh.
>
> I could narrow down to these but that is still a larger combination.
>
> ENABLED_SERVICES="n-api,n-cpu,n-net,n-vnc,openstackx,quantum,q-agt"
>

On the second node we need n-cpu and q-agt

Dave Lapsley from Nicira recently did a presentation on multi-node devstack
installation and that might help answer your questions

http://www.slideshare.net/delapsley1/opensack-quantum-devstack-tutorial

From a Quantum perspectiv, the OpenVswitch plugin documentation should
indicate the required flags -
http://openvswitch.org/openstack/documentation/
A link to this website should be in the README for OpenVswitch plugin.

>
>
> (3)
>
> Assume you have two Compute Nodes connected on 2 Ports of an Upstream
> Access Switch.
>
> We create a network L2 N1, and launch Guest 1 on Host 1 and Guest 2 on
> Host 2.
>
> So will the two Guests G1 and G2 be on same L2 Network N1, or we have to
> create a seperate L2 Network N2 for G2.
>

Quantum networks are "virtual networks" therefore if L2 N1 is a Quantum
network,
both guests will be on the same network. This network can be a private
network
that belongs to a tenant or shared network accessible by all Nova tenants (
also known
as projects)

>
> How will L2 connectivity work between G1 and G2 in regards to switch
> ports? How does Physical Switch know VLAN Tags for N1 to allow Trunk
> mode? Or they will be internal to Hosts and not span Vswitches?
>

Every packet leaving the vSwitch will be tagged with the correct VLAN that
represents
the "Quantum L2 Network"

>
> How will L3 connectivity work in regards to NAT and routing for VMs?
> Think the Nova Network on Node 1 would be acting as Router/Default
> Gatway for Node 2, and do NAT, but it brings into dual NIC topology and
> without switch direct connect.
>

Your assumption is correct.

>
> I was thinking of all Compute Nodes being connected to Switch with one
> NIC.
>
> How does two Vswitches talk with a Switch in middle
>

The two vswitches just tag teh packets with appropriate VLAN and forward it
to the physical L2 fabric to deliver the packet to the other vSwitch.

>
>
> (2) For Floating IP
>
> http://docs.openstack.org/diablo/openstack-compute/admin/content
> /associating-public-ip.html<http://docs.openstack.org/diablo/openstack-compute/admin/content/associating-public-ip.html>
>
> Will this be Ok example to follow?
>

yes, but Quantum started supporting Floating IP as of Essex-3

>
> I was not sure if G1 and G2 could share a L3 Subnet on different nodes.

> Here a topology diagram becomes very important as there are many ways to
> connect L2/L3.
>

As Quantum gets is absorbed into OpenStack Core for Folsom, we will try to
put
more documentation around reference physical network design. Uptill now we
have
been focussed on OpenStack side and haven't put much thought into
recommending
the physical network architecture that works best. Although Rackspace has a
OpenStack
reference architecture that you might find helpful -
http://www.referencearchitecture.org/

>
> --
> You received this question notification because you are an answer
> contact for quantum.
>

--
Somik Behera | Nicira Networks, Inc. | <email address hidden> <email address hidden> |
office: 650-390-6790 | cell: 512-577-6645