Re: Multihoming Issues

[Simon Leinen <[e-mail]>]

On Tue, 03 Sep 2002 19:51:23 -0700, David Conrad <[e-mail]> said:
> On 9/1/02 7:30 AM, "Simon Leinen" <[e-mail]> wrote:
>>>> - one prefix for each ISP in the world
>>>> - one prefix for each POP or campus in your network
>>>> - one prefix for each LAN in your POP or Campus
>>>> - additional prefixes that you decide to carry for your own reasons (eg,
>>>> policy)
>>> My, that's a lot of prefixes.  I'm sure I'm missing something here.
>> Probably - note how the scope gets narrower as you go down to smaller
>> parts of the Internet.

> Well, yeah, but if you want to gain full benefit of multi-homing,
> each of these prefixes would need global visibility, no?

If you want to enjoy the FULL benefits of multi-homing, all the other
folks you want to community with must be multi-homed too, so that
there are no single points of failure :-)

Probably the typical site could achieve 95% of the benefits of
multihoming with an impact on only 5% of the global Internet.

What is needed is some sort of feedback loop that weighs the interest
of multi-homing entities against its impact on remote parts of the
infrastructure.

In the spirit of "think globally, act locally", here's what we do as a
regional ISP: We have relatively strict inbound route filters based on
RIR assignment policy and traditional defaults for swamp space (/24 or
shorter in 192.0.0.0/7, /19 or shorter in the non-RIR-administered
part of 0.0.0.0/2 and 128.0.0.0/3 etc.).

In our neck of the woods, most ISPs send e-mails to their peers when
they want to start announcing new paths.  When I notice that the of
the new prefixes violate our prefix filters, I respond that we won't
be accepting those routes by default, but would agree to add an
exception to the filters for a limited amount of time if the new
customer will renumber into a larger aggregate.

This at least creates a small incentive for ISPs to think about
RFC2260-like solutions for multihoming customers, or for customers who
(are forced to) change ISPs to renumber into their new ISP's space.  A
caveat is that we have partial routing, carrying mostly only European
routes, so we don't drop traffic to networks whose routes we filter.
-- 
Simon.