Date: Fri, 14 Jun 2002 14:23:26 -0400
From: Margaret Wasserman <[EMAIL PROTECTED]>
Message-ID: <[EMAIL PROTECTED]>
| Even if a multi-sited host has no routing protocols and a primitive
| forwarding table (i.e. an ICMP redirect cache), there will still be
| increased implementation complexity. The multi-sited host will
| need to maintain multiple forwarding tables (one for the global
| scope, and one for each attached site), will need to choose the
| correct table as part of the next-hop lookup for outbound traffic,
Yes, and it also needs to do that for link local addressing as well.
That's the point I am trying to make - not that site locals are free,
but that they're not all that expensive given that almost all the same
mechanism is required for link locals anyway. The insides of the
routing processes are the one major exception (as link locals never
participate in those).
| and will need to make sure that ICMP redirects are applied to the
| correct table when received.
I'm not sure it isn't possible in some obscure setups to get redirects
for link locals, which would require just the same - but in any case,
the interface (which is a more specific form of the scope) needs to be
checked when processing redirects in all cases, to make sure that a
redirect is coming from a router on the link that you're sending the
packets to for the destination involved (the redirect will be from a
link local addr, which should normally be the same one as the LL
addr of the next hop for the route, and the LL addr has scope already,
and is a subset of a site scope).
| And what do we actually get for this added complexity?
I think this has been answered enough. People are ignoring the
benefit as no-one is actually being made to renumber their IPv6
sites currently.
When Bob Fink made a suggestion on the 6bone list to change the
6bone addresses (the prefix length actually) I suggested also taking
back all the current assigned prefixes, and reassigning them using the
new rules, just so everyone (using 3ffe::/15 anyway) would actually
experience renumbering their nets - from which we could perhaps learn
something.
But people are still pretending (hoping) that no-one is ever going to
have to renumber their nets, and so don't want to actually test any of
this part of IPv6. The desire to do away with SL is based upon much
the same misconception. The costs are obvious to implementors, and
the benefits are currently invisible.
kre
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page: http://playground.sun.com/ipng
FTP archive: ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------
