[11004] in Commercialization & Privatization of the Internet
Re: California NAP Designed as a CIX Killer??
daemon@ATHENA.MIT.EDU (Peter S. Ford)
Thu Mar 17 22:36:17 1994
From: "Peter S. Ford" <peter@goshawk.lanl.gov>
To: cook@path.net (Gordon Cook)
Cc: hwb@upeksa.sdsc.edu, com-priv@psi.com
In-Reply-To: Your message of Wed, 16 Mar 94 20:21:02 +0000.
Date: Thu, 17 Mar 94 15:05:51 MST
>>> Peter - are you saying that those who connect to a NAP may choose to peer
>>> (do I use the right term?) only with whatever subset of NAP connectees
>>> they choose?
Gordon, you are using the right term. There can be technical
reasons why you do not want to coerce networks to peer. For example,
2 NSPs connected to a NAP might have some other exchange point where they
are exchanging traffic (peering). They should have the flexibility to
choose how they want to peer. Also, it is possible you might only
want to peer at only a few of the places where you could possibly
peer with another NSP, thereby reducing the total amount of routing
information in your system.
There are advantages to many NSPs peering at the "same spot" as can
be evidenced by the existing practices at the CIX and MAE-East.
It makes thinking about the system more tractable, it simplifies
debugging the system, it simplifies the routing strategies used by
NSPs, it makes it easier to plan deployment, it makes it easier for
users to grok what their pings and traceroutes are telling them, etc.
It also makes it easier to plan and engineer inter-domain backup in the
event of NSP failure.
peter
