[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: multi-homing vs multi-connecting



    > From: RJ Atkinson <rja@extremenetworks.com>

    >> Multihoming makes routing more complex, as it adds more possible paths.

    > That is true of current IPv4 multi-homing, when talking about
    > inter-domain routing. It is NOT necessarily true for all approaches to
    > IPv6 multi-homing. In fact, it might be preferable to select an approach
    > that minimises the complexity increase on inter-domain routing due to
    > multi-homing.

I think we probably ought to distinguish between several different kinds of
"the routing is involved in this multi-homing solution".

On the one extreme, there's the "this thing has only one routing-name, and the
normal path-selection algorithm is always generating routes to it through
currently working links" - in other words, the current IPv4 multi-homing model.

But there are numerous others, e.g. Dave Clark's old "route fragments" idea
(and sorry, I don't have a citation for it), where along with the destination
"address", you get back a number of source route fragments, leading from
"large" topology elements your local router probably does know about (e.g.
major international ISP's) to the destination via various access paths to that
destination.


It's important to realize that, fundamentally, multi-homing (of *all* kinds)
*is* just a path-selection - i.e. routing - problem. I.e. there's more than
one way to get to the destination host, and somehow everyone/thing involved
has to get together to i) work them out, and ii) select one.

The problem is that the straightforward ways of doing this, ones that work in
a small network, don't work in something the size of the Internet because the
routing overhead explodes. So then you start trying to reduce the amount of
work; perhaps by offloading some of the work in various ways (some of which do
not look like routing).

Sure, there can be other non-routing implications, e.g. if you decide that the
you're going to use is multiple "addresses", then you have to deal with the
"how do I tell that address A and address B refer to the same host, the I'm
talking to" (provided of course that that problem is one you care about).

But basically multi-homing turns into a path-selection (i.e. routing) problem.


	Noel