Re: Transport level multihoming

[Daniel Senie <dts@senie.com>]

Thomas Narten wrote:
> 
> Daniel Senie <dts@senie.com> writes:
> 
> > OK. So today end users can multi-home using certain NAT boxes, having
> > zero cost to upgrade to IPv6 a fixed cost to get multihoming, and no
> > recurring costs.
> 
> Can you please clarify what you mean here? My understanding is that
> traditional NATs allow a site to avoid needing to renumber, but don't
> help with multihoming.

There are multi-ported NAT boxes (e.g. Radware Linkproof) which permit a
site which uses NAT to have two or port connections to the Internet. The
connections are load balanced among the connections, can handle failure
of one link, etc. If you can live with NAT, it's a way to handle
multihoming after a fashion. You do get to specify your inbound policy,
too, since traffic sent out a particular link will come back that way.

> 
> Or are you referring to a specific NAT implementation that allows
> sharing of state across boxes, and if so, do the NAT boxes use
> different external IP addresses (i.e., to different providers), and
> does (say) a TCP connection that somehow migrates from one box to
> another continue to work transparently?
> 
> That is, what are the details on how this works?

This is something that's done entirely in one box (with the potential
for a hot spare in the case of the vendor mentioned above). There can be
many links upstream of the box.

If you like, I could go into more detail about how such a product could
be designed. I am aware of two products on the market using this type of
approach. I don't have product details beyond marketing materials on
either. I do know how to build such a box, though, from previous work
I've done in the NAT space. It's not all that difficult to do the load
spreading, though the failed link detection and any performance
optimizations (e.g. use the link which has a shorter path to a
particular server) allow for all kinds of interesting differentiators in
products.

Personally, I'd rather recommend customers use IPv6 and a decent
multihoming solution provided therewith. If IPv6 multihoming is no
different than IPv4 multihoming, and companies can't realistically get a
good multihoming solution, I'm afraid they're likely to ignore IPv6 and
go with a NAT solution which permits multihoming, and call it a day.


-- 
-----------------------------------------------------------------
Daniel Senie                                        dts@senie.com
Amaranth Networks Inc.                    http://www.amaranth.com