lkml.org 
[lkml]   [1997]   [Jan]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Know tcp bug ?!
Alan Cox wrote:
>
> > I have problem with squid and setsockopt with SO_RCVBUF options.
> > This is a dump that show that tcp is not OK. Note the window size
> > of 700, that is the option that I use in squid.Is this a known bug ?
>
> If you set your SO_RCVBUF below twice the MSS of a TCP socket then
> its behaviour is not going to be good.
>
> > This is strange. Is send an ack with win 0. (There have not been any
> > data, so it should be the value I have put in to the
> > setsockopt value.)
>
> The setsockopt value is the maximum memory committed to the socket NOT
> the window size. Consider the case of a daemon that wants 500 connections
> all to use small amounts of memory. If I create 500 clients and send all
> 1 byte data blocks that 700 bytes will actually take in the region of
> 64Kbytes.
>
> Setting the window using SO_RCVBUF is the wrong way to do it. Instead
> set the window on routes you need to force a low window with something
> like
>
> route -net add 44.0.0.0 netmask 255.0.0.0 gw 44.1.2.3 window 700
>
> Alan

Im not sure that I understand it. I do understand when you say that it
not a good way, and the route is the rigth way. But the way I try to do
it gives my an advatage to control it in the applikation. ( By the way,
do not what to do it. I trying do se how thing behave...)
I do not understand way it not should work. And I you for sure could say
'this can't work' should the setsockopt return a error value in that
case ?

\
 
 \ /
  Last update: 2005-03-22 13:38    [W:0.043 / U:0.364 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site