Re: Virtualization And IPv6 : Part Of IPv6 Address For Virtual Hosts

[Mark Smith <ipng at 69706e6720323030352d30312d31340a.nosense.org>]

On Wed, 26 Sep 2007 10:38:11 +0100
<michael.dillon at bt.com> wrote:

> > Alternatively, you could create a virtual network/link within 
> > the physical host, 
> 
> This is precisely what machine virtualization already does. And many
> setups involve a set of physical hosts providing resources, and a set of
> virtual machines consuming resources. A management system migrates the
> virtual machines from host to host as it tries to make the best use of
> the resources. At any given point in time, the only way to truly know
> what virtual machines are on a given physical host is to ask the
> physical host.
> 
> Since the virtual machines use virtual network interfaces provided by
> the physical host, this physical host can see MAC addresses and IP
> addresses.
> 

I'm guessing that the virtual hosts/virtual network is bridged on to
the physical network, so the external physical hosts and the virtual
hosts are layer 2 adjacent. Where I think my suggestion is different
was the idea that this virtual network is routed "onto" the physical
segment, meaning that a link local multicast on the virtual network
would only be seen by the virtual hosts.

I think the bridging idea is a better one because it avoids introducing
a layer 3 hop. However if you need to reduce layer 2 neighbour
adjacencies on your physical / virtual network, or need the link-local
multicast only reaches peer virtual hosts, I think setting up a virtual
router would be a reasonable way to do it.

> As a management problem, this can be easily solved by an agent which
> runs on the physical hosts therefore I don't think the IETF needs to do
> work on this problem. What would be good is for more people to implement
> pure IPv6 networks on their virtual machine infrastructure and write
> about it.
> 
> Since virtualization increases the number of IP addresses consumed per
> physical machine, it could lead to IPv4 exhaustion happening sooner
> rather than later. By using pure IPv6 on the virtual machines there is
> effectively no limit to the number of addresses that can be used, and if
> someone wants to implement some kind of structured numbering system in
> their /64 subnet, the bits are freely available to do this.
> 
> This USENIX paper:
> http://www.usenix.org/events/usenix06/tech/menon/menon_html/index.html
> Optimizing Network Virtualization in Xen
> provides some information on how one of the more popular virtualization
> environments handles networking.
> 

Thanks for the ref, I'll have a look.

Regards,
Mark.

> --Michael Dillon
> 
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> ipv6 at ietf.org
> Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------

--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6 at ietf.org
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------