From: Radharamanan Radhakrishnan Newsgroups: comp.parallel.mpi,comp.parallel.pvm Subject: Re: Poor MPI performance with collection of Linux workstations Date: 17 Mar 1999 09:47:28 -0500 Organization: University of Cincinnati, ECE/CS News Server Message-Id: References: <36ED325A.69D8C9B8@rrsg.ee.uct.ac.za> <7cn7up$4tq$1@grandprime.binc.net> <7cnhh8$aig$1@grandprime.binc.net> Xref: ukc comp.parallel.mpi:4767 comp.parallel.pvm:8135 > discovered that there is a funky Linux tcp problem with sending small > chunks of data that persists with the new kernels. hi, In response to your observation, i have the following addendum. This was posted in the linux newsgroups sometime back that talks about the same problem. Here is the text of that conversation: ------- Start of forwarded message ------- FYI: An explanation on how a simple patch can boost Linux TCP bandwidth by a factor of 20 for short messages (short means <500 bytes in kernel 2.0.36, <100 bytes in kernel 2.2.2). http://www.icase.edu/coral/LinuxTCP.html The performance fix involves turning off delayed ACKs on highly interactive links where TCP_NODELAY socket option is desirable. This is particularly applicable to parallel applications using MPI. Sincerely, Josip -- Dr. Josip Loncaric, Senior Staff Scientist mailto:josip@icase.edu ICASE, Mail Stop 403 http://www.icase.edu/~josip/ NASA Langley Research Center mailto:j.loncaric@larc.nasa.gov Hampton, VA 23681-2199, USA Tel. +1 757 864-2192 Fax +1 757 864-6134 ------- End of forwarded message ------- hope this helps ... cheers, ramanan -- Research Assistant, Computer Architecture Design Laboratory, Department of Electrical and Computer Engineering and Computer Science University of Cincinnati, 805C Rhodes Hall, Cincinnati, OH 45221 Lab-Ph : 513-556-0904 URL: http://www.ececs.uc.edu/~ramanan