The CIX and the NSFNET regionals - a dilemma


How technology can solve the problem:


  If you mark packets with a indication of whether they are commercial or
R&E, and then build a router which understands the indication, then you can
route each packet accordingly (and also solve the mixed-R&E case nicely.)

I'd point out (probably unnecessarily) that if you replaced "commercial"
and "R&E" with "wants maximum reliability" and "wants maximum throughput"
it would immediately conjure up a way to maintain a split routing table,
with type-of-service. While no current routers support this, there are
at least bits in the IP header to be used for this, and OSPF could carry
the routes and get the dual routing tables right if someone actually
bothered to implement that part of the protocol.

Now, if you had routers which had a knob which said "for any packet
which arrives though this interface with the default TOS, change the
TOS to one of commercial or R&E", in this way allowing you to classify
packets as "AUP-compliant" or "commercial" when they entered your network,
you could effectively do what was suggested above, all with existing
but seldom-implemented bits of the protocol. It also might allow users
to pick their own preferred routing by setting the TOS bits on their

This isn't perfect. It isn't quite clear to me how one would
assemble the dual routing tables, picking the outbound direction
properly for networks which can only be routed one way while splitting,
networks with alternative routing, without importing all routes from
everywhere. Despite this, however, it seems like about the quickest
way to obtain source-based policy routing in cases where only a limited
number of policies (like two) exist, since most of the protocol knobs
exist already.

Dennis Ferguson