[498] in Commercialization & Privatization of the Internet
Re: CIX
daemon@ATHENA.MIT.EDU (Dennis Ferguson)
Fri Mar 29 20:54:39 1991
From: Dennis Ferguson <dennis@utcs.utoronto.ca>
To: geoff@fernwood.mpk.ca.us
Cc: com-priv@psi.com
Date: Fri, 29 Mar 1991 20:52:41 -0500
> Not exactly. PSI (and I'd venture to guess AlterNet and ANS?) have third
> party traffic prohibitions in their agreements. The NSF use agreement has
> no mention of third party traffic prohibitions. This is what allows all
> those UUCP, PhoneNet, et al sites to hang off NSFNet regional network site
> connections and enjoy the spoils of National Networking Now on the gov't
> dole. (This is also why policy based routing is a fairy tale -- unless you
> are policy base routing based on end-to-end IP connectivity you're
> operating in fantasy land. Say a site with a local area network with
> workstations that hangs off a regional net member site with UUCP is not
> going to be subject to policy based routing based on the traffic's true
> origin, rather it will be based on the origin of the site that the UUCP
> site calls into).
I think your opinion of policy-based routing derives from a definition
of "policy-based routing" which not everyone would agree with. I think
that the IETF's IDPR group, as I understand its work, is attempting to
address the types of situations you mention. This is "policy-based routing"
and probably isn't a fairy tale, though it is a hard problem and hard
problems have hard solutions which may not have universal appeal.
The confusion comes from the fact that "policy-based routing" (which
probably qualifies as a winner in the "phrases with confused definitions"
category) is sometimes used to refer to something which is qualitatively
different. In particular, Merit uses this term to describe something
I prefer to call "route selection", or "route preference", refering to
the process of deciding which, if any, of the routes to a network being
advertised to you by multiple neighbours should be used, and deciding
what information about networks you know routes to should be revealed
to your neighbours.
Route selection deals with trying to make sane use of complicated
hop-by-hop network topologies, but punts on usage policy issues except
at the coarsest level. BGP, for example, advances the state of the
route selection art by providing additional topology information from
which routing decisions can be crafted, but doesn't really provide
much in the way of additional help in dealing with realistic usage
constraint problems. Indeed the few, primitive mechanisms of usage
policy enforcement we use now, based mostly on network number
recognition, will be even less effective on an OSI Internet where
routing information is coalesced into objects which represent larger
clusters of organizations and individuals, some "appropriate" and
some not.
I think the original assertion, that some think that policy-based
routing is the solution to the usage policy problem, may indeed be
correct. It's just that policy-based routing isn't something
that is currently being done, and any belief that it is now being
done really derive from confusion over definitions.
Dennis Ferguson
