Re: (ipv6mh) the Rebel Alliance meetings in Atlanta (long)

[Iljitsch van Beijnum <iljitsch@muada.com>]

On Tue, 3 Dec 2002, Kurt Erik Lindqvist wrote:

> (I trimmed the CC list...)

Good, I got some of your messages three times...

> > X per ISP is not going to work as there are global ISPs and very, very
> > local ones.

> Well, what I meant was that the day an ISP run out of their PI blocks
> they will just continue down the address block, just as ISPs today are
> announcing more specifics than their RIR allocation and complaining
> that these are not accepted. I don't see why we think people would
> behave differently just because this is IPv6....

We think that because they do. At the moment, at least. Also, we want
them to, as current IPv4 routing practices aren't exactly great. On the
one hand it is good that people want to keep the IPv6 routing table
clean, on the other hand we don't want to be so restrictive it doesn't
work. Remember that most people have a good reason to pollute the IPv4
routing table. Just not allowing this without offering alternatives
isn't good enough.

> > It would be much, much better if we could agree that everyone does this
> > the same way. Having a route in 90% of the routers isn't much better
> > than having it in 100% of the routers, but being able to reach 90% of
> > the internet is much worse than being able to reach 100%.

> That is the way it works for many (perhaps most) of the multihomers
> today. 90% is better than the 0% we have today.

Disagree. People who multihome care about their connectivity more than
others. For PI, 90% isn't enough. 99% isn't even enough. On the other
hand, for shooting holes in PA, having a 90% backup is probably good
enough, but I'd rather have 99% or even 100%, especially if I have to
eat the cost and annoyance of renumbering when changing my primary ISP.

> Still, I agree that if
> we can come up with a common solution that would be better. My proposal
> is to accept /48 for the time being and not scale back until there is a
> new solution.

I could live with that if we can get consensus among operators. Without
those /48s in the routing table this is still too risky as it doesn't
protect you against an entire ISP going down.

However, I don't think it's the best solution.

Also, if we end up with a multi-address solution, it could very well be
that all PA will be tied to interconnect locations. For instance, I live
between two of the largest exchange points in the world: the LINX and
the AMS-IX. If I were a single homed multi-address capable network I
would want addresses that are always routed over the LINX and also
addresses that are always routed over the AMS-IX, so I can do traffic
engineering.

I'm not saying we should start doing this now, but it would be good to
anticipate future developments.

> > If geo aggregation isn't feasible, then I think a fixed block of PI /48s
> > for each RIR would be the next best option. I think that everyone can
> > agree that 4 or 5 x 1024 - 8192 (for a total of 4096 - 40960) routes
> > wouldn't be too bad if we can keep a tight lid on this. And if and when
> > other blocks are commisioned, people can choose to carry or filter
> > those, without hurting existing multihomers.

> What is it that we think we gain with these specific blocks, be it for
> site-locals or multihoming? The number of routes are not going to be
> less.

This way you can easily identify these routes with a prefix filter. I'm
not saying that's always absolutely necessary, but as long as we have
the choice, why not create this option?

> If the RIRs use up their current assignments -good! Then at least
> we have a IPv6 network from where we can start drawing conclusions.

Fully agree.

Iljitsch

BTW, I'm writing an article about if/when IPv6 will be adopted for a
Dutch magazine. I'm still looking for good quotes from IPv6-skeptics.  :-)