Skip to content
Snippets Groups Projects
  1. Jan 01, 2019
  2. Dec 31, 2018
  3. Dec 30, 2018
  4. Dec 29, 2018
  5. Dec 28, 2018
  6. Dec 25, 2018
  7. Dec 24, 2018
  8. Dec 22, 2018
  9. Dec 21, 2018
  10. Dec 20, 2018
  11. Dec 19, 2018
    • rswindell's avatar
      Important update to the default TCP socket values defined in this file: · c5e7bdfb
      rswindell authored
      - no longer override the default send and receive socket buffers (limiting
        to 8Kbytes) - apparently modern TCP/IP stacks can use much larger buffers
        for sending large data more effeciently (quickly) over fast networks.
        Eliminating these 2 lines sped up the web server (e.g. wiki.synchro.net)
        page loads by a factor of 20+.
      - enable TCP_NODELAY (disabling the Nagle algorithm) for *all* TCP sockets
        by default (not just telnet and rlogin servers).
        This helps performance of large data transfers a little bit too.
      
      Big thanks to Deuce (!) for helping to debug and identify the cause of the
      poor performance of the web server. I'm not sure why I/we were overriding
      the default send/receive buffers but the old sockopts.cfg file introduced
      in 2003 had 4Kbyte buffer sizes specified, and then when we switched to
      sockopts.ini in 2005, the buffer sizes were increased to 8Kbytes. This might
      have been the defaults for say, the Windows TCP/IP stack at the time, I
      forget where I got those numbers from. Anyway, we shouldn't override them
      without a good reason. And today, we definitely don't have a good reason
      that I'm aware of. The default buffer sizes perform *much* better!
      c5e7bdfb
  12. Dec 18, 2018
  13. Dec 17, 2018
Loading