RE: RE : [lwip-users] what was the concensus on the tcp send problem

[Julian Gardner [RSD]]

could not use the last cvs as it hung on netconn_delete, just tried new version 
and working a lot better thanks.


Julian Gardner

RSD Communications Ltd
8 BorrowMeadow Road
Springkerse Industrial Estate
Stirling, Scotland FK7 7UW

Tel   +44 [0]1786 450572 Ext 238 (Main)
Fax   +44 [0]1786 474653
GSM   +44 [0]7836 240907

Email address@hidden
Skype joolzg
MSN   address@hidden
WWW   http://www.rsd.tv   http://www.rsdtv.com





-----Original Message-----
From: address@hidden on behalf of Goldschmidt Simon
Sent: Wed 25/07/2007 18:49
To: address@hidden
Subject: RE : [lwip-users] what was the concensus on the tcp send problem
 
>> If i send a block of data of say 4K the first block is sent but there
>> is a large delay sending the 2nd and 3rd block.
>
> Even with the patch (already check in the CVS HEAD)  from
> https://savannah.nongnu.org/bugs/?20506
> <https://savannah.nongnu.org/bugs/?20506>  ?

The patch raises the initial congestion window from 1mss to 2mss. The initial 
congestion window is responsible for this delay (tcp slow start algorithm). 
Until now, it was set to the conservative (and deprecated?) default of allowing 
only one segment of maximum size to be transmitted without receiving an ack, 
regardless of the window size it adverts or you set it to. Since the initial 
window size is now set to 2 mss and most other stacks will send an ack after 
receiving 2 frames (delayed ack), this should be done now. Simply upgrade to 
CVS HEAD if you can.

Simon

<<winmail.dat>>