Date: Tue, 23 Jul 1996 19:03:08 -0400
To: salo@msc.edu (Tim Salo)
From: Paul Ferguson <pferguso@cisco.com>
Subject: Re: sell shell accounts?
Cc: freedman@netaxs.com, nanog@merit.edu>The desire for a full mesh of VCs between routers becomes more compelling
>if you have more than three routers. The diagram is left as an exercise
>to the reader; I rather dislike having to draw in ASCII.
>
>Note that the two paths between A and C, (directly versus via B), seem
>to take distinctly different paths. (Perhaps, I missed your point).Perhaps I missed your point.
Tim, I could've sworn that you had been following the IETF ION WG
mailing list, where this has been discussed on hundreds of occasions.
I hesitate to even mention this on NANOG (I really don't want to
re-hash this on this list), but I can't hold my tongue.The scalability of an ATM network where each end-point is fully meshed
to all other end-points decreases in direct proportion to the number
of end-points. This is not so much a layer 2 problem, per se, but rather
the ability to maintain functional routing (convergence, et al.) in a
fully meshed network topology.This does not scale.
I believe that given my [secret] assumptions, (some of which are revealed
below), it scales well beyond the financial abilities of most any NSP.
I assumed that the wide-area ATM service supported only PVCs and
PVPs. (I keep hearing promises of SVCs "next year" but I don't think
anyone [domestically] has delivered them.) I believe this address your
concern about maintaining functional routing, but I make no claims to
being a routing expert, so I might not fully comprehend the issue.
I also assumed that the going price for a DS-3 xwide-area ATM connection
is, to more or less pull a number out of the air, $250,000/year.
This has two implications:
o Any one NSP isn't going to be able to afford all that many
connections. A full mesh of 100 routers, (i.e., $25 millino/year
in ATM services) would require 4,950 PVCs. That is a lot to
configure, but...
o Given that each link is rather expensive, it behooves the
NSP to try to avoid needlessly moving a packet back and
forth across a DS-3 local loop just because my ATM
provider can't provide me with any more PVCs.
(Which was my major point, by the way.)
On the other hand, you are correct that this doesn't sale to thousands
of routers. But, I didn't think that was the problem we were
trying to solve here, (just yet).
-tjs