Re: solution to NAT and multihoming

[Jon Crowcroft <J.Crowcroft at CS.UCL.AC.UK>]

In message <3298.980237371 at cs.ucl.ac.uk>, Jon Crowcroft typed:

 >>if multihoming is killing routing coz default free zone routers have
 >>too many entries
 >>and NAT is killing users coz they can't get always on addresses
 >>why not have multihomed sites (aren't they usually server/core
 >>provider sites) LEASE their standby link address prefixes to access provider
 >>sites - and swap the address prefixes when their default link fails
 >>and they need to failover to the standby link/addresses...
 
 >>symmetry dictates this ought to work out...and everyone wins

 >>by setting uo as a market we could even make the incentives right...

i wasn't too clear about this (a bit like my lousy 1000 bit error in
the port nat message - that'll teach me to send emails before i've had
any coffee:-)


so after suitable basting by sean doran, here's the scoop:-

I like GSE; however we dont have v6 and we do have NATs; we also have
multihoming.

1/ consider global DHCP as a tool, and a mechanism for buying a
lease on an aggregate

2/ do NATting on aggregates

3/ design a BGP attribute (yech, i know) to inidcate that an address
range is "bank switchable" - this means that it is part of a lease
from one AS to another. This means that when told (via management, BGP
update, or designated "important" ingress or egress link failure), a
pair of domains then bank switch the address range, but enable NATing
on the range for exsting flows...

got it?

j.