On 21-apr-2006, at 15:47, Brian E Carpenter wrote:
If we abolish that myth and look at the problem we are left with
an answer where BGP passing regional aggregates is sufficient.
I'm sorry, I don't think I've ever seen a convincing argument how such
aggregates could come to pass in the real world where inter-regional
bandwidth is partitioned at link level or dark fibre level. There just
isn'tany forcing function by which mathematically close prefixes
will become topologically close, because there's nothing that
forces multiple providers to share long-distance routes.
Obviously it would be tremendously helpful if ISP A would handle
region X, ISP B region Y and ISP C region Z, so it's just a matter of
dumping the traffic for a certain region with the ISP in question but
for various reasons this will never work, if only because the whole
point is that multihomers have more than one ISP and each of their
connections to their ISPs may be down at any point in time.
Let me try out a new analogy. There are many languages in the world,
and most international businesses have to be able to work in more
than one language. Now wouldn't it suck if in a business that has
customers in 25 countries with 20 languages, EVERY office would have
to have people that speak EVERY language? Fortunately, although there
is no fixed relationship between language and geography, in practice
there is enough correlation so that you can have the office in Sweden
handle all Swedish speaking customers and the offices in Portugal and
Brazil the Portugese speaking customers.
Back to networking: send the packets to the place where all the more
specifics are known. If the place where all the more specifics are
known is close to the places where those more specifics are used,
that works out quite well. If the more specifics are used randomly
all over the place then this technique adds detours, which is
suboptimal.
The point that Scott was
making is that there are proposals for non-random assignments
which could be
carried as explicit's now and aggregated later.
I understand his point very well and I'm even in favour of it, because
statistically it can only help aggregation and certainly not damage
it.
But I think that without some radically new principle it will at best
very slightly reduce randomness.
I guess I'll work on my simulations...
_______________________________________________
Ietf@xxxxxxxx
https://www1.ietf.org/mailman/listinfo/ietf