[11306] in Commercialization & Privatization of the Internet
All CIX members are equal redux
daemon@ATHENA.MIT.EDU (Sean Doran)
Mon Mar 28 19:46:18 1994
To: com-priv@psi.com
Date: Mon, 28 Mar 1994 09:49:25 -0800
From: Sean Doran <smd@cesium.clock.org>
I wrote yesterday:
One of the most interesting things about the CIX, and what, IMHO,
makes it a decent DMZ, is that everyone connects to it on an exactly
equal footing.
This wasn't correct. There is one exception to the rule
that all directly-connected CIX members are on an exactly
equal footing.
ANS CO+RE's line to the CIX is treated uniquely; while it has
the same bandwidth as the other lines, it connects to a different
router, which does lots and lots of route filtering.
This was done originally because the way ANS does it's routing
allowed any regional attached to ANS that took a default route
(essentially all of them) to reach any CIX route that was configured
in the MERIT PRDB as exiting ANS to AS 1957.
Essentially, this meant that not only ANS CO+RE itself got
routing via the CIX, but so did the CO+RE resellers that
weren't CIX members and also all the R&E-only regional networks.
This was raised here some time ago when someone in Canada
complained that he could no longer get packets between
CA*Net and an AlterNet no-NSFNet (CIX only) network. That was
shortly after the IGS was set up to do filtering on the ANS
CO+RE connection to the CIX.
I believe that the CIX offered an official reason that involved
a legal question of whether or not traffic between a CIX-only
network (whose administrators never agreed to abide by the
NSFNet AUP) and an R&E-only network (which has connectivity only
via the NSFNet backbone) was acceptable.
The principal question that raises is one of who is responsible
for such traffic, especially when a gross violation of AUP
happens (and one was seen here, in fact). It would not surprise
me to learn that the CIX board received advice to the effect
that the CIX would have some liability in such cases.
There is also an obvious non-legal issue, and that involves
what has been argued back and forth on com-priv for the last
several days, viz. the CIX loses out economically when ANS
CO+RE's gateway customers suddenly get free CIX routing while
SprintLink's equivalents have to join the CIX and pay $10000.
The legal question is a hot one, but could be solved fairly quickly if
the NSF stated categorically that traffic between a network that
has not agreed to the NSFNet AUP and one that has no non-NSFNet access
is acceptable, and that the CIX Association and its individual members
bear no liability whatsoever for any AUP violations as a result of that
traffic. Alternatively, ANS could block all traffic between non-CO+RE
(i.e., NSFNet-only) networks and the CIX. Finally, ANS could adjust
its routing system so that it allowed only itself, its direct
customers, and its gateway customers who are also CIX members to route
traffic to CIX-WEST.
The third alternative is obviously what the CIX board would like
to see, and I believe that the CIX board's opinion is both
broadly representative of the CIX membership's and in the
membership's best interests.
However, the first two alternatives serve to reduce the problem
from one of a complex legal question about AUP liability to one
of an argument between the CIX board and one CIX member about
policy and the interpretation of the CIX by-laws.
Sean.
