The problem here is on the Solaris end of things. It's just a bit to
anxious to retransmit things. The main problems seem to be
in the algorithm for estimating round trip times, and in the
algorithm to do slow startup on a TCP link. You can help your
performance a bit by picking a smaller value for MRU
on your PPP link. I find performance on a 14.4k link
gets acceptable with an MRU of 552. I'd be interested to
hear what happens with a 28.8 connection. Note that you still
need to transfer at least 60k or so of data before the
cost of the unneeded retransmits starts to be swamped by
the amount of data you are sending.
I'm not totally convinced that I've done everything I can to fix
this, but its looking like there is very little more to be done,
at least from our end of things. As a data point, does anyone
know of systems that get good performance interacting with a
Solaris box over a 14.4 or 28.8k PPP or SLIP link with large
MRU/MTU settings? I'd be very interested in seeing packet
traces of fast sessions with a Solaris box over such a link.
-- eric
---------------------------------------------------------------------------
Eric Schenk www: http://www.cs.toronto.edu/~schenk
Department of Computer Science email: schenk@cs.toronto.edu
University of Toronto