Date: Mon, 10 Feb 1997 13:25:00 -0500 From: Garrett Wollman <wollman@lcs.mit.edu> To: Greg Burch <gregb@qosnet.com> Cc: freebsd-current@freefall.freebsd.org, Christoph Kukulies <kuku@gilberto.physik.rwth-aachen.de> Subject: Re: tcp/ip performance miserable Message-ID: <9702101825.AA05237@halloran-eldar.lcs.mit.edu> In-Reply-To: <32FF5A7A.41C67EA6@qosnet.com> References: <199702092032.VAA14356@gilberto.physik.rwth-aachen.de> <32FF5A7A.41C67EA6@qosnet.com>
next in thread | previous in thread | raw e-mail | index | archive | help
<<On Mon, 10 Feb 1997 12:27:22 -0500, Greg Burch <gregb@qosnet.com> said: > I am also seeing very slow performance with 3.0-current. Not only ftp > but also nfs (using tcp) and reading messages from a pop server. My > system is a Pentium 166 and SMC PCI EtherPower NIC. I do not have a > problem when running this same hardware with R2.1.5 For people who have performance problems with TCP, I would strongly suggest doing a graphical performance analysis with tcpdump and xplot. It is much easier to figure out problems when one has anaylzed what is happening. For the record, I get pretty close to the expected value on the two tests I was able to do from 3.0(current) to Digital UNIX 3.2G and an old 2.2 SNAP. The graphical performance analysis technique is described in a technical report, MIT/LCS/TR-494, by Timothy J. Shepard, entitled ``TCP Packet Trace Analysis''. (It was originally a Masters' thesis.) The `xplot' program described in the report, along with a Perl script I wrote to generate xplot input files from tcpdump output, and maybe also a PostScript version of the TR, is available at <ftp://mercury.lcs.mit.edu/pub/shep/xplot-0.89.tar.gz>; I should make a port of it some day. The usefulness of this technique cannot be overstated. It is literally the difference between poring over text traces for three hours and looking at a plot for five minutes and saying ``aha, there's a fast-retransmit, and there's where another packet got dropped, so that delay is a retransmit timeout!'' It's like night and day. -GAWollman -- Garrett A. Wollman | O Siem / We are all family / O Siem / We're all the same wollman@lcs.mit.edu | O Siem / The fires of freedom Opinions not those of| Dance in the burning flame MIT, LCS, ANA, or NSA| - Susan Aglukark and Chad Irschick
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?9702101825.AA05237>