[1532] in Commercialization & Privatization of the Internet
Re: Alternate Routing
daemon@ATHENA.MIT.EDU (Deborah Estrin)
Thu Oct 24 16:49:53 1991
Date: Thu, 24 Oct 91 13:42:54 PDT
From: estrin@usc.edu (Deborah Estrin)
To: steve@ncri.cise.nsf.gov
Cc: yakov@watson.ibm.com, com-priv@psi.com
In-Reply-To: Stephen Wolff's message of Tue, 22 Oct 91 08:43:22 EDT <9110221244.AA29790@cise.cise.nsf.gov>
Reply-To: estrin@usc.edu
I am in strong agreement with Steve on this (to no one's surprise I
expect) although I think Yakovs points are well taken and need to be
kept in mind.
For any small number of widely used classifications you can
accommodate the small multiplier in the size of your routing table and
just use the equivalent of TOS bits.
And in fact, if you identify a small n number of service types and/or
small n number of user classifications AND instances of those types
exist in large numbers (majority) of domains, then it pays to go ahead
and precompute and store the n routing entries and you dont need to
think of it as source-specific routing.
HOWEVER, for those classifications that are not widely represented it
is simply inefficient/impractical to precompute and store the alternatives in
all border routers that such packets may encounter. For this scenario
you do source-demand installation of source-specific routes.
So, eventually we need to accommodate both ways of establishing a
forwarding path for packets. What you do with any particular case at a
particular point in time depends on the policy/service requirements
and the distribution of the affected traffic.
D.
