[8980] in North American Network Operators' Group
Re: UUNET Pulling Peering Agreements & replacing them with charging under non-disclosure?
daemon@ATHENA.MIT.EDU (Peter)
Fri May 2 11:09:14 1997
Date: Fri, 02 May 1997 11:02:35 -0400
From: Peter <peter@tdi.net>
Reply-To: peter@tdi.net
To: Jeremiah Kristal <jeremiah@corp.idt.net>
CC: nanog@merit.edu, inet-access@earth.com
Jeremiah Kristal wrote:
>
> Sure, if UUNET was only cutting peering with small ISPs who were only at
> one NAP, and had peering because of a backdoor deal years ago. It appears
> that UUNET is cutting peering with those medium-sized ISPs who *have*
> built a national DS-3/OC-3 backbone. I really don't see how squeezing the
> medium-sized ISPs who have already invested the millions of dollars it
> takes to build a national backbone helps anyone but UUNET.
> I see this as a direct attack on the smaller regional and national ISPs
> who have been taking customers away from UUNET because of better
> performance and better service.
I see this as driven by practicality and economics.
- Eight or so "national backbones" have materialized in the last year.
Each
one of those requires that engineering devise and then implement
a routing policy, each of which is different than all the others
written before because of a different combination of exchange points
the new net is present at and how well things are going at the various
EPs at any given time. When some of these "national backbones" are
advertising a total of 5 or six nets, how practical is this? It's
easier to static route them and forget about it.
- BGP peering sessions cost money. Few people seem to understand this.
All
routers from all manufacturers have some practical limit to how many
BGP sessions they can support while simultaneously forwarding packets.
(In fact, one manufacturer recommends that you use a router solely as
a router server to maintain sessions and feed forwarding info to
other routers in order to allow them to concentrate on forwarding
packets.)
Every time you have to put in a new router to manage BGP sessions
that don't generate revenue and don't forward packets, you have to
try and figure out if it's economically worthwhile.
- The exchange points have become impractical for the largest players;
it's
just too hard to diagnose performance problems between two nets when
there's a piece neither of them controls in between. Thus we have
private interconnects. Private interconnects are circuitwise
inexpensive
when the two nets are in the same POP already, but not when actual
local loops are involved. Also, a private interconnect requires a
high speed port which costs real money on each end. You want to get
your money's worth out of each one of those ports, which means it
needs fairly high utilitization to be justified; a DS3 port
moving 2Mpbs is not justifiable. So network
interconnection is an issue of both practicality and economics.
- So, in order to provide the best service possible to their customers,
the largest nets have gone to private interconnects amongst themselves.
This means that they now have these expensive connections to
the various exchange points which they maintain in order to
move a small percentage of their total exchange traffic. Again,
this can appear economically questionable. Every one of these
nets wants to become and stay profitable, and the way to do it
is to reduce unnecessary fixed/recurring costs whereever possible.
People who've not directly worked on or managed a heavily loaded net
may not appreciate everything I've said.
-peter