Spdtest - test your raw TCP socket performance
2011-05-30 Version
Michael Brutman (mbbrutman@gmail.com)


Introduction

  This program can be used to test the speed of your machine when
  sending and receiving data on TCP sockets.  It has no practical
  use other than for benchmarking so don't invest too much time
  with it.

  The basic idea is to send or receive data using the TCP library
  without doing any extra processing on the data.  The data will
  not be copied, written to disk, or manipulated in any way.
  It will be fully processed by the TCP library though, just like
  normal data - no shortcuts like not computing the checksums are
  taken.  This allows you to determine the maximum speed you should
  expect when sending or receiving data.  Actual applications will be
  slower because they presumably do work.

  I use this program to test the speed of new versions of the mTCP
  code.  I also use it to figure out the relative speed of different
  machines and Ethernet cards.


Hardware requirements

  8088 processor or better
  190KB available RAM
  CGA, Monochrome (MDA), EGA or VGA display
  Supported Ethernet card, SLIP or PPP connection


Software requirements

  DOS 2.1 or newer (DOS 3.3 or newer recommended)
  Packet driver for your Ethernet card, SLIP or PPP connection


Setup instructions

  Spdtest is built using the mTCP library.  The setup instructions for
  mTCP can be found in SETUP.TXT.


Using Spdtest

  Spdtest uses the following syntax:

    spdtest -target <ipaddr> <port> [options]
    spdtest -listen <port> [options]

  The first form of the command tells spdtest to create a socket
  connection to another machine.  The second form of the command
  is used to make spdtest wait for an incoming connection from
  another machine.


  Options are:

    -help             Show basic help text
    -srcport <port>   Use <port> as the source port when connecting
    -receive          Do a receive only test
    -send             Do a send only test
    -mb <n>           Megabytes to send during a send test


  Spdtest requires another machine to either send it data or to blast
  data at.  To ensure a good benchmark the other machine should be
  far faster than the machine you are testing.  For example, I use a
  200Mhz Linux machine to test my slower PCs against.

  The other machine should be running netcat.  If spdtest is doing a
  send test then the other machine should be set to receive and throw
  away the data as quickly as possible.  If spdtest is doing a receive
  test then the other machine should be sending a fixed amount of data,
  hopefully as fast as it can.


Tips for getting accurate results

  - Both machines should be on the same network
  - The network should be relatively quiet
  - The machines should be plugged into a switch, not a hub
  - The test should run at least 10 seconds, and possibly up to a minute.
    The longer you run the more accurate the results will be.
  - Timing should be done with a stopwatch and on the other machine.
    Some Ethernet card and packet driver combinations hold off interrupts
    too long, distorting time on the machine running spdtest.
  - Check the ending messages for network errors.  The test should run
    cleanly; network errors are a sign of other problems.


Support

  Have a comment or need help?  Please email me at mbbrutman@gmail.com.


More information: http://www.brutman.com/mTCP

Created May 20th, 2011, Last updated May 30th, 2011
(C)opyright Michael B. Brutman, mbbrutman@gmail.com

