constrained socket buffers patch

[coderman]

On 7/13/07, Juliusz Chroboczek <Juliusz.Chroboczek at pps.jussieu.fr> wrote:
> ...
> There is another issue, which doesn't appear in the above.  TCP is
> extremely sensitive to packet loss when the window is smaller than
> 4 packets (fast retransmit doesn't work in that case).  So could
> I suggest a default value for ConstrainedSockSize of 8 kB?

this is a good idea.  16k might be even better if it worked reliably
(the usual default is 32 to 64k).

i've updated the patch to provide better documentation on this option
in the man page and set the default to 8k per your suggestion.[0]

one last note: this should work on win32 sockets and may help reduce
the windows buffer problem. [1]

any feedback testing this patch on vservers or win32 is appreciated.

best regards,


0. updated constrained socket buffers patches:
  http://peertech.org/pub/tor-constrained-sockets-0.1.2.14.patch
  md5:ff2432af49e3348e12de61a795b10840
  sha1:8a18a8a1b2dc872fff03d7611773f615b3382d28
  ---
  http://peertech.org/pub/tor-constrained-sockets-0.2.0.2-alpha.patch
  md5:b53e37863612110e764769b84f527609
  sha1:7e990582ea0a240bd54b3ba7ea97ede72a592856

1. "Windows Buffer Problems (WSAENOBUFS)"
  http://wiki.noreply.org/noreply/TheOnionRouter/WindowsBufferProblems