Fw: [ppml] Re: ARIN Policy Proposal 2002-7

[Shane Kerr]

On 2002-10-08 11:38:50 -0400, Charles Scott wrote:
> 
> While I admit to being more management these days than technical, I'm
> curious why there isn't a technical solution to multi-homing with long
> prefixes. It would seem that some well placed systems could provide
> aggrigation of a number of long prefix attachments and provide routing
> service to those long prefix customers via tunnels. Such a service
> would provide a solution to impact on the global routing tables, offer
> ultimate portability to the small customer who needs to be multihomed
> but has a small IP footprint, and perhaps could provide a revenue
> stream to whomever built it. In fact, haven't I seen such offerings
> already available?

In my understanding, a route is the combination of address + AS, and
that eventually leads back to a single router:  the origin.  The origin
is a single point of failure, period.  Ideally this will be the place
with the highest reliability - in most cases this is in an ISP, but end
users may see it differently (I know my personal network's reliability
is bounded largely by my ISP, rather than myself).  Multihoming in the
small, as it were, allows users to control this for themselves.

Kind of like the end-to-end principle?  Not a perfect analogy, I
suppose.  :)

Anyway, if I understand your proposal then this solves various kinds of
single points of failure, but still leaves end users relying on a single
ISP for service.


There is a interesting draft for the IETF working group looking at
solving the IPv6 multihoming issue, which covers IPv4 multihoming in
some detail:

http://www.ietf.org/internet-drafts/draft-ietf-multi6-multihoming-requirements-03.txt

Since policy is being set based on limiting routing table growth, does
it make sense for ARIN to expend resources in this area?


My personal belief is that the best solution to this problem is to use
something like SCTP for a transport.  SCTP uses multiple IP addresses in
a single connection.  This would allow the elimination of all single
points of failure within the network, and also remove the motivation for
end sites to multihome.  Obligatory link:

http://tdrwww.exp-math.uni-essen.de/inhalt/forschung/sctp_fb/

-- 
Shane                      If you choose the quick and easy path [...] 
Carpe Diem                           you will become an agent of evil.
                        - Yoda, lecturing Luke on Software Engineering