Xen users - Jul 2011 - Binding Xen vm migration traffic to travel over 1 network card (nic)

We are running Xen 4.0 on Debian Squeeze. We have set up a 2-node cluster and
successfuly migrate vm instances between them.

But we want to be able to select a particular nic to handle the migration
traffic. Xen documentation and discussion sometimes refers to a hypothetical
network configuration in which ''one network interface goes to the
outside world, and the other, connected by crossover cable, handles xen
management, including migration.''

Right, so how is this setup achieved, such that migration travels only over one
nic on each of the 2 nodes, the xend xm commands having been bound to travel
over that nic?

An example command, assuming box1 and box2 are the 2 nodes:

box1# xm migrate lucidxentest.phon box2 --live

To be clear, I am pretty sure this is not a vm (domU) configuration issue;
rather a dom0 configuration issue.

Is the binding to a particular nic achieved through Xend configuration? or is it
done externally, by adding routes, etc.?
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Benjamin Weaver wrote:
>Is the binding to a particular nic achieved through Xend 
>configuration? or is it done externally, by adding routes, etc.?
OK, first off it''s not a Xen issue, rather a general IP routing issue 
- the key is understanding how kernel routing works.

There are several ways to achieve what you are looking for.

The first, and probably simplest is to have a separate network for 
management traffic. Add a second NIC in each physical machine, and 
hook them all together either via a switch, or for two machines, with 
a cable.
Use a private network address on the management network, and address 
management traffic to the private addresses. Eg, if you give hsot1 
the IP address 192.168.57.1, and the second 192.168.57.2, then you 
could simply route management traffic via that network by using the 
appropriate address.

Ie, if on host1 you send traffic to 192.168.57.2, then the kernel 
will automatically route it via the management network since it''s a 
directly connected network.

As I said, this is by far the simplest and needs no manual 
intervention to routing tables. It works because by default, the 
kernel will always route traffic for "locally connected" hosts via a 
directly connected network - ie if a host is in the same subnet, then 
it can be reached without going through any routers. Only if the 
destination isn''t directly connected will it start looking for other 
routes - most usually a single default router.


If you want to still refer to hosts by their "frontend" 
addresses/URLs, then you could setup the same backend/management 
network, but you''d need to put explicit routing rules in. It''s
outside my area of expertise, but I would have thought you should be 
able to put in a static host route (on host1) so that a.b.c.d routes 
via 192.168.57.2, and the corresponding reverse route on host2 of 
a.b.c.e routes via 192.168.57.1. I''m not sure if you need to do 
anything special to make this route take priority over the directly 
connected route via the frontend network.

There are probably some more "imaginative" and complicated 
techniques, but I suspect the first will do you.

-- 
Simon Hobson

Visit http://www.magpiesnestpublishing.co.uk/ for books by acclaimed
author Gladys Hobson. Novels - poetry - short stories - ideal as
Christmas stocking fillers. Some available as e-books.

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users