Re: [v6ops] State of play as of today

[Mark Smith <[e-mail]>]

Hi,

On 2 October 2015 at 08:56, Brian E Carpenter
<[e-mail]> wrote:

<snip>

>>>  Certainly, this work could be discussed including the fact whether the consensus is to remain within the 64-bit boundary between the interface-id vs. not for any IPv6 addressing.
>
> I would strongly suggest discussing draft-smith-enhance-vne-with-ipv6 at
> the same time.
>

To save people a bit of time, I did the following presentation on that
draft at the AusNOG 2014 conference:

"Network Virtualisation: The Killer App for IPv6?"
http://www.slideshare.net/MarkSmith214/nvtkaipv6-50785685


Tom asked me to review his ILA draft a while back. In the context of
both his and my drafts, and having had some further time to think
about them, I think there might be two different but related problems
being considered:

- network virtualization, which I think might actually be more
accurately described as "network emulation"

- virtual machine / process portability across physical machines and
grouping of them together so they share a prefix for policy and other
reasons


My draft was purely focussed on the first problem, although it
inherently addresses the second. From memory, I think Tom's was
specifically addressing both of the problems.

The reason I've come to realise there are probably two different
problems to consider is that one of the fundamental differences
between what I proposed (and what other network
virtualization/emulation methods propose) and what Tom proposed is
that in his model, the hosts within the virtual networks have to have
IPv6 addresses that are structured to suit the operation of the
underlay network. This is easy or easier to achieve when the operator
of both the underlay and overlay networks is the same
entity/organisation.

I think that difference is fairly significant from the point of view
of trying to perform network emulation successfully. It seems to me
that if you're emulating a network, then your fundamental goal is that
the emulation of the network should be as transparent as possible to
the hosts using the emulated network, which includes allowing them to
use what ever addressing / addressing structure they'd like.

I think Tom's proposal is a good solution to the second problem,
however I wouldn't see it as a complete network emulation solution
because of its client host addressing requirements. I do think there
is a need for solutions to both types of "network virtualization".


Regards,
Mark.