Re: /24 multihoming issue

[John Payne <john () sackheads org>]

On Oct 20, 2005, at 2:07 PM, Randy Bush wrote:

> > > > Is 7018 preferring 19094 over 701 regardless of
> > > > AS-PATH length?
> > > the convention is that, if 19094 is a customer of
> > > 7018, then it will always prefer it.
>
> and it was confirmed that this is the case for the
> prefix in question
>
>
> > And this is a good reason not to cross "tiers" of your
> > transit providers.  Either have both "transit free" or
> > both should have transit.
>
> why?  when it get up to tier-1s it will be the same, the
> one(s) who heard it from customers will prefer the
> customers.
>
> and tier-Ns should be preferring customer routes as well;
> see discussion here between vaf, asp, and me in about '96.

slipping back into the tier terminology which i was trying to avoid...

it's only a problem if you want to do inbound traffic engineering.   
If the tier 2 is well connected to tier 1s (for example Internap),  
it's typically going to get more inbound traffic than the tier 1  
connection because the tier 2 is preferred as a customer in a bunch  
of tier 1s.