RE: recent slowdown in routing table growth

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

On Thu, 9 Jan 2003, Tony Li wrote:

> |   Why do people keep blaming multihomers for the mess that is
> |   the global routing table?

> Because it is an unsolved technical problem that needs to be addressed.
> Please consider the very long term.  What will IPv6 look like in 2050?

Long term concern is reasonable, but today's mess has nothing to do with
multihoming.

> |   Based on what we can see today, this is utter nonsense.
> |   There are maybe 10000 multihomers in the world.

> So?  It will continue to grow.  I can remember when the default free
> table was 5000 routes.  Total.  And the number of multihomers will
> eventually start to grow faster than the number of ISPs.  We're
> already seeing ISP contraction.

But this doesn't mean consolidation of the prefixes they announce so
this is largely meaningless. In IPv4, the problem is unsolvable.
Unfortunately, IPv6 is inheriting too many of the problematic IPv6
policies but the larger address space and stricter rules should help.

> And today those 10000 cost more resources to the DFZ than any other
> 10000 sites.

Today 50 incompetent network administrators cost more resources than the
10000 most resource-costly but well-managed sites.

> We need an architecture where people can multihome without a DFZ routing
> table entry.  And yes, we need to fix a lot of other problems too.  But
> any one of these problems could eventually eat us alive and we have to
> fix all of them.  My vote is that we not forget the harder technical
> issues and leave the social engineering issues for those that are more
> qualified.

We all know what needs to be done but somehow that's not enough to make
it happen.

There are basically two ways to engineer solutions to problems that are
hard to solve by conventional methods:

1. Use conventional/proven/obvious solutions and simply keep working
   until you have something that works well enough to be usable
2. Think out of the box

The problem with type 1 solutions is that they tend to be huge and
complex and this only gets worse over time. The problem with type 2
solutions is that they often seem ridiculous. "Packet switching? That
will never work." "Reduce the instruction set? That's nonsense."
"Publish the encryption algorithm? Don't be silly." It turned out that
packet switching, RISC CPUs and public scruteny of encryption algorithms
were major steps forward.