[e2e] TCP slow start and maximum segment size

Mark Allman mallman at grc.nasa.gov
Thu Jul 24 12:42:50 PDT 2003

(clearly I cannot keep up with email; sorry for the slow response)

> I checked all the RFCs mentioned as well as RFC 1122,
> RFC 2001 but none of them give a value for slow start
> threshold.

RFC 2581 notes that the initial value of ssthresh may be arbitrarily
high (e.g., the receiver's advertised window).  I think this is the
right approach in most cases(*) (i.e., let TCP slow start until it
finds the congestion point of the network or the receiver's limit).
Using some magic default constant will certainly help in some cases
and hurt in others and so I think in general the right approach is
to figure it out on the fly.

* My comment is for current, standard TCP.  Sally has a riff on slow
  start for HighSpeed TCP that makes good sense to me (i.e., when
  the cwnd gets to some size doubling it in 1~RTT makes me a bit


Mark Allman -- BBN/NASA GRC -- http://roland.grc.nasa.gov/~mallman/

More information about the end2end-interest mailing list