Re: TcpAgent::bug_fix_

[Sally Floyd <floyd@ee.lbl.gov>]

...
>if one wants to run the
>protocol as it "usually" runs, should bug_fix_ be on or off?

Good question.  "Bug_fix_" is a recommended fix to a bug in the
underlying Tahoe/Reno protocols.  There is currently a discussion
in the tcpimpl working group about how "bug_fix_" should or should
not be folded in to the IETF's current revision of RFC 2001.  As
far as I know, some TCP implementations have it, and some don't.
(Just as some current TCP implementations now have SACK, and
some don't.  Like everything else about the Internet,
a characterization of how TCP "usually" runs is very much
a moving target.)

My own philosophy has been that NS is intended to model
the underlying dynamics of TCP's congestion control algorithms,
and not the reality of some particular current implementation of
TCP.  Thus, the one-way Reno implementation does not include the
well-documented bug in BSD Reno's congestion control algorithms
of a multiplicative factor in the window increase algorithms.
The two-way Reno TCP implementation in NS has options, turned off
be default, that allow behavior closer to that actually in 4.4 BSD.

- Sally