As far as I can tell, trying to create an IPv6 address that is
mathematically equivalent to the AS number does us very little good if
we assume some sort of mapping system (like ILNP.) It also does us very
little good if we assume that customers use PA addresses.
First, note that AS numbers are now 32 bits. So an AS number solution
presume IPv6. (That doesn't actually bother me, but it is worth
mentioning.)
Second, the reason I referred to an Ipv6 address is that I am not going
to assume global replacement of core routers and global changes in the
core forwarding paradigm. (Edge changes, yes, we have discussed many of
those.)
Given that, then the primary goal we have already been working towards
is that the IPv6 addresses used in packets in the core be aggregatable
addresses. As such, if we got to the point that these were the only
addresses that were used for packets in the core, we would see
significant benefit in advertisement / FIB size.
Folks could argue that operators today disaggregate such blocks. And
indeed they sometimes do. They do so for good reasons. Attempting to
tell the operators that they MUST use a system that does not permit that
(such as a pure AS based system without the ability to circulate
multiple advertisements for the AS, will mean that they will refuse to
adopt the solution. For example, while I do not agree that the
operators must have exclusive control over traffic engineering (which is
an argument that has been used against shim6, LISP, ILNP, ...), it is
true that they should have control of the traffic engineering of the
paths to their own prefixes.
Yes, I would prefer to address such a requirement by different tools
than disaggregation and path stuffing. previous efforts to go there
have foundered on teh lack of an incentive for the core operators to
make any changes in that space. (They don't get any benefit until
almost everyone has shifted, and they incur MAJOR costs.)
And if we had a routing system that gave us TE without disaggregation,
then PA address blocks would be a small enough multiplier that we
wouldn't care if each operator advertised a few of them.
Hence, there may be some computational (internal to the router) reasons
to focus on the AS first, and there are some dynamics reasons to wonder
about using an AS-centric approach, as long as we have heavily
disaggregated advertisements, in practice this is not in my view really
a solution to the problem.
Yours,
Joel
Paul Jakma wrote:
On Sat, 8 May 2010, Dae Young KIM wrote:
It seems your idea is based on global IP addressing. What I have in
mind is to get rid of global addressing.
Yes, I gathered that, but it's unnecessary AFAICT.
So, IDR routers look only at AS numbers, not at IP addresses which
bears no meaning in IDR.
Sure, but why? :) What problem are you trying to address? If you're
trying to force aggregation by AS, then why not use an IPv6 prefix that
embeds the ASN? Basically, the underlying forwarding technology isn't
the problem - the problem is finding a good way to apply some
higher-level structure to it.
IPv6 almost certainly has the room to accomodate any reasonable
addressing structure that can't fit with IPv4. If you're proposing a
solution that requires all forwarding infrastructure to be upgraded,
it'll have to have amazing benefits, and be impossible to layer onto IP.
Also you need to consider "Why do we currently have ASes that originate
multiple disparate prefixes?".
regards,
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
