Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 29 Nov 2008 12:39:49 -0800 (PST)
From:      Venkat Venkatsubra <venkatvenkatsubra@yahoo.com>
To:        David Malone <dwmalone@maths.tcd.ie>
Cc:        Rui Paulo <rpaulo@fnop.net>, freebsd-net@freebsd.org, Kevin Oberman <oberman@es.net>
Subject:   Re: FreeBSD Window updates
Message-ID:  <10716.94191.qm@web58304.mail.re3.yahoo.com>
References:  <200811291746.aa88825@walton.maths.tcd.ie>

next in thread | previous in thread | raw e-mail | index | archive | help
Hi David,=0A=0AAssuming each of these 5 window updates are sent after the a=
pps=0Acompleted the=A0read that it issued, what's the buffer size the apps =
is passing=0Ato the read ? If it is small, could that be increased ?=0A=0AB=
ut I agree with you that there is room for optimizing sending of these wind=
ow updates.=0A=0A(Since I don't know the window scale factor for the connec=
tion, I couldn't figure out=0Ahow much each of the window updates was openi=
ng up the window.)=0A=0AVenkat=0A=0A=0A________________________________=0AF=
rom: David Malone <dwmalone@maths.tcd.ie>=0ATo: Venkat Venkatsubra <venkatv=
enkatsubra@yahoo.com>=0ACc: Rui Paulo <rpaulo@fnop.net>; freebsd-net@freebs=
d.org; Kevin Oberman <oberman@es.net>=0ASent: Saturday, November 29, 2008 1=
1:46:30 AM=0ASubject: Re: FreeBSD Window updates=0A=0AI've got an example e=
xtract tcpdump of this at the end of the mail=0A- here 6 ACKs are sent, 5 o=
f which are pure window updates and=0Aseveral are 2us apart!=0A=0AI think t=
he easy option is to delete the code that generates explicit=0Awindow updat=
es if the window moves by 2*MSS. We then should be doing=0Asomething simila=
r to Linux. The other easy alternative would be to=0Aadd a sysclt that lets=
 us generate an window update every N*MSS and=0Aby default set it to someth=
ing big, like 10 or 100. That should=0Aeffectively eliminate the updates du=
ring bulk data transfer, but=0Amay still generate some window updates after=
 a loss.=0A=0ANormal ACKing for driving congestion control shouldn't be imp=
acted=0Aby either of these suggested changes.=0A=0A=A0=A0=A0 David.=0A=0A12=
27622713.276609 172.16.2.2.5002 > 172.16.1.51.39077: . ack 144097745 win 40=
798 <nop,nop,timestamp 5365425 5103763> (DF)=0A1227622713.276611 172.16.2.2=
.5002 > 172.16.1.51.39077: . ack 144097745 win 40830 <nop,nop,timestamp 536=
5425 5103763> (DF)=0A1227622713.276613 172.16.2.2.5002 > 172.16.1.51.39077:=
 . ack 144097745 win 40862 <nop,nop,timestamp 5365425 5103763> (DF)=0A12276=
22713.276615 172.16.2.2.5002 > 172.16.1.51.39077: . ack 144097745 win 40894=
 <nop,nop,timestamp 5365425 5103763> (DF)=0A1227622713.276852 172.16.2.2.50=
02 > 172.16.1.51.39077: . ack 144097745 win 40926 <nop,nop,timestamp 536542=
5 5103763> (DF)=0A1227622713.276855 172.16.2.2.5002 > 172.16.1.51.39077: . =
ack 144097745 win 40958 <nop,nop,timestamp 5365425 5103763> (DF)=0A12276227=
13.296585 172.16.2.2.5002 > 172.16.1.51.39077: . ack 144100641 win 40956 <n=
op,nop,timestamp 5365445 5103766> (DF)=0A__________________________________=
_____________=0Afreebsd-net@freebsd.org mailing list=0Ahttp://lists.freebsd=
.org/mailman/listinfo/freebsd-net=0ATo unsubscribe, send any mail to "freeb=
sd-net-unsubscribe@freebsd.org"=0A=0A=0A=0A      



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?10716.94191.qm>