From: Chris L. <cla...@re...> - 2006-09-27 19:35:05
|
Sridhar Samudrala wrote: > Chris Lalancette wrote: > >> Hey all, >> I'm kind of new to SCTP, so forgive me if I ask silly questions. >> I am looking at the latest upstream lksctp implementation (I pulled >> Linus' git tree this morning). I am running iperf with a minor patch >> to make it use SCTP for testing (I'll attach details about the patch >> below). When I set up iperf to use buffers of length 65536, I am >> seeing some on-the-wire behavior that I don't quite understand. >> Basically, the connection starts as normal: >> >> INIT >> INIT_ACK >> COOKIE_ECHO >> COOKIE_ACK >> >> After that, the sender sends the first data packet, which in this case >> has CTSN of 520344188. The receiver sends a SACK for 520344188. Then >> the sender sends two more data chunks, with CTSN of 520344189 and >> 190. The receiver then sends a SACK for CTSN 520344190. Up to here, >> things seem fine. After this is when I don't understand. The next >> two packets are SACK's from the receiver, changing the a_rwnd (which I >> believe is allowed by the protocol), but with CTSN of 520344191. How >> can the receiver be SACK'ing a CTSN that it hasn't yet received? Even >> more strange is that the next data packet that the sender sends is for >> CTSN 520344192, and as far as I can tell (with searching through >> ethereal), CTSN 520344191 is *never* sent. >> > > It is possible for ethereal to drop some packets under heavy load. > You must be running into that situation. > > Thanks > Sridhar > >> Am I missing something obvious protocol-wise here, or is this a bug? >> All of the relevant information is here: >> >> download/compile iperf with sctp: >> http://people.redhat.com/clalance/sctp/sctp_iperf.sh >> ethereal packet capture: >> http://people.redhat.com/clalance/sctp/2.6.18.cap >> >> Thanks for your help, >> Chris Lalancette >> >> ------------------------------------------------------------------------- >> Take Surveys. Earn Cash. Influence the Future of IT >> Join SourceForge.net's Techsay panel and you'll get the chance to >> share your >> opinions on IT & business topics through brief surveys -- and earn cash >> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV >> _______________________________________________ >> Lksctp-developers mailing list >> Lks...@li... >> https://lists.sourceforge.net/lists/listinfo/lksctp-developers >> > > > Sridhar, Grr. I didn't know that about ethereal. Thanks for the info. Chris Lalancette |