I have been trying to track down some slowness in the tcp stack when the
connection is over a non-perfect ethernet connection.  In my situation it
is a thinnet network.  I have had phenomenal problems, with a large range
of kernels.  In order to try to speed things up a little, I even put a
switch on one end of the connection to clean up bad packets, but it didn't
help.  Where a Windows box might get 800+KB/sec, my linux box is lucky to
get 5KB-30KB/sec.  I have the same results whether I use ftp, netcat, or
anything else.  The client in this case is running 2.1.125 or 2.1.131ac7,
doesn't matter.  The host can be any number of machines, but the
performance is usually the slowest when connecting to another linux box. 

I had problems with older kernels where the transfer would run very
quickly, and the receive window size would drop to 0.  The transfer would
then drop to a very slow rate and never recover.  The most recent dumps
that I've done never have the receive window go away, but there are a lot
of cases where there is an almost .5 second delay at times before an ack
occurs. 

I would be willing to do any testing that could possibly help fix these
bugs- they are a major problem for all of my linux boxes that are not
solely on switched ethernet.

I have captured some tcpdump output from both ftp and netcat, which I can
make available via ftp for anyone who could use them.


Brian Pape
Computer Resource Services
University California Los Angeles
[EMAIL PROTECTED]
voice: (310) 825-9284
fax: (310) 206-6039


-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]

Reply via email to