The CIX and the NSFNET regionals - a dilemma

John Curran jcurran at
Wed Feb 5 03:18:37 UTC 1992

   I recently wrote a breif message on this same topic which mentions
some possible solutions (none good).  I attached it here to help clarify
the issues involved for those network service providers which have a major
research component.



Date:     Mon, 27 Jan 92 22:22:13 EST
From:     John Curran <jcurran at>
To:	  Folks:;
Subject:  NSFNET and CIX routing issues

Hello All,
  John Rugo passed along the message regarding the various routing
issues that arise for a regional that connects to both the CIX and the NSFNET.
I would like to summarize the situation that occurs to insure that we are all
working on the same problem.  We welcome any and all thoughts on this issue.

Given that a basically research and educational network (such as NEARnet) has
a high-speed connection (assume 10 or 45 Mb/s) to the NSFNET and also connects
to CIX to exchange commercial traffic, here are the ramifications:

   Any network which is only CIX-reachable presents no problems.  You accept
   the route for the site and send any traffic via the CIX.  Routes to all of 
   NEARnet's members are made available to that site and their network sends 
   traffic for NEARnet members via the CIX.

   Any network which is NSF AUP compliant and prefers to use the NSFNET is not
   a problem:  You insure that such network routes  are not announced via the
   CIX and NEARnet members will automatically use the NSFNET to reach them.
   Likewise, NEARnet will not announce routes to these "R&E" networks via CIX,
   which prevents any return traffic from taking the wrong path.

Note that both of the above policies work, but cannot be simultaneously 
implemented in almost any network today.   The policies would require that
routes back to an R&E site (say, "Harvard") be sent to the CIX so that a 
purely commercial site (say, "TRW") can access them, yet it also requires
that the route to Harvard not be announced in order that any other CIX-
reachable R&E site (say, "Stanford") does not utilize the slower path.

This is only a problem if:

  - There exists R&E members that utilize more peak bandwidth than the
    CIX path will allow.  (There is: Harvard Astrophysics, MIT Laboratory
    for Computer Science, BBN packet video, etc..)

  - The routes that are used by R&E sites is the same as the routes used by
    your non-R&E sites. (They are: packets are generally routed based on 
    destination, and not on any history such as "color" or source network)
  - There exists both R&E and non-R&E traffic on the network (There will be;
    or there is no reason to pursue the CIX).

How technology can solve the problem:

  If you can separate your network into two distinct networks with separate
routers for R&E and non-R&E members, then you can maintain distinct routing
tables for each type of member.  This is not pretty, or cost effective.

  If you place a box at your border which not only routes on destination but 
also on the source of the packet, then you can effectively maintain distinct
routing tables for R&E and non-R&E sites.  This does create a traffic 
bottleneck and single point of failure.

  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.)

That's the situation to the best of my knowledge.  If anyone sees a solution,
or has questions about the assumptions, I'd be interested in hearing from you.


------- End of Forwarded Message

More information about the NANOG mailing list