On 04/20/06 at 1:54pm +0200, Brian E Carpenter <brc@zurich.ibm.com> wrote:
Scott Leibrand wrote:
...
If you don't have source address rewriting, and we assume that ingress
filters are in place (which seems the by default assumption) then the
source address selection actually determines the exit path form the
multihomed site.
Not in today's networks. Currently routers route on destination address,
not source address, so the destination address selection determines which
route is used to route the packets, which in turn determines which source
address you have to use if you want your packets to get out.
Actually, the sending host could theoretically use a different default gateway
for each different source address. That could be used as a back door method
for choice of exit router, without needing to touch router behaviour.
True, for a small network, where all hosts are on the same LAN as the exit
routers. I was referring to a larger network with an IGP.
This seems to me like yet another way that shim6 is designed for small
sites, and is somewhat less suitable for larger ones. This isn't
necessarily a bad thing, as IMO host-based multihoming makes less and less
sense the larger your site gets, but it seems to be a factor a lot of
people don't want to accept.
If we target shim6 at small-site multihoming, for networks that can't do
BGP-based multihoming today (or would prefer not to for cost), I think we
have a better shot at getting something actually deployed. If networks
like Igor's and Patrick's aren't going to benefit from shim6 without
significant additional state in their servers, they're going to stick with
BGP multihoming. If we recognize that, and work to make sure that his
servers can talk shim6 to multihomed clients with minimal overhead, we'll
be less likely to get caught in the catch-22 of being unable to get
traction because one end isn't interested in deploying shim6.