PAIX (was Re: Interconnects)

Majdi S. Abbas msa at samurai.sfo.dead-dog.com
Sun May 19 02:54:32 UTC 2002


On Sat, May 18, 2002 at 04:51:27PM -0400, Ralph Doncaster wrote:
> One BGP session instead of dozens is more convenient.  Maybe not more
> useful for engineering, but certainly less work than negotiating and
> configuring a bunch of sessions for bilateral peering.
> 
> For smaller ISPs like mine, knowing in advance that you won't get snubbed
> for peering after connecting to an exchange is the big attraction.  Given
> the dozens of signatories on the AADS MLPA, it looks like they can be
> quite popular.

	Strictly speaking, I don't think a route-server is required to
multilaterally peer, but they certainly help.  However, there are a couple
of big catches, particularly on an ATM or similar switching fabric:

	1) One or two sessions, one or two VCs...if they go down, you will
	lose all your peering at that site.

	2) The possibility of blackholing traffic to a peer who you have
	a downed VC to, but who is still advertising their prefixes to 
	the route server.

	Additionally, quality of peering does not necessarily correlate
to quantity of peering.  I'm not going to claim that it's a bad thing 
to peer with a large number of typically smaller providers, but they
don't always account for a statistically signifigant portion of your
traffic.  If you're going to have to negotiate bilateral agreements to
cover the bulk of your peering traffic, why not consistantly negotiate
bilateral agreements?

	--msa



More information about the NANOG mailing list