[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Re: constrained socket buffers patch

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

Your information might be somewhat obsolete...

Have a look at my machine (a pretty ordinary recent Linux) connecting
to tor.eff.org:

  lanthane.45747 > SWE 4264190125:4264190125(0) win 5840 <mss 1460,sackOK,timestamp 315382960 0,nop,wscale 6> > lanthane.45747: S 3004159902:3004159902(0) ack 4264190126 win 65535 <mss 1460,nop,wscale 1,nop,nop,timestamp 1445924711 315382960,sackOK,eol>

Lanthane (my machine) opened a connection to tor.eff.org.  Since it
doesn't have any information about the connection's RTT yet, it
declares a very small TCP window (5.8 kB).  Tor.eff.org has a somewhat
more traditional implementation of TCP, so it starts with a window
size of 64 kB.

(Note that both sides negociated the wscale option, so from now on we
need to multiply lanthane's values by 64, and tor.eff.org's by 2).

A little while later, the situation has evolved as so: > lanthane.pps.jussieu.fr.45747: P 4112:4445(333) ack 4629 win 33304 <nop,nop,timestamp 1445925772 315383225>
lanthane.pps.jussieu.fr.45747 > . ack 4445 win 408 <nop,nop,timestamp 315383277 1445925772>

As you may see, tor.eff.org has grown its receive window to 67 kB,
while Lanthane has gone up to 26 kB.