Date: Wed, 28 May 2003 12:52:39 -0400 From: "kfl" <kfl@xiphos.ca> To: "freebsd - net" <freebsd-net@FreeBSD.org> Subject: TCP for transaction Message-ID: <JCEDLMKGMLLELHAJNIHJGEGHEJAA.kfl@xiphos.ca>
next in thread | raw e-mail | index | archive | help
Here is a tcpdump output from one FreeBSD box with TCP for transaction enabled and a client supporting the NO_PUSH setsockopt with a windows box: 19:00:15.357281 192.168.3.200.1247 > 192.168.2.7.microsoft-ds: SP 4208572194:420 8572331(137) win 65535 <mss 1460,nop,wscale 2,nop,nop,ccnew 2> (DF) 19:00:15.357514 192.168.2.7.microsoft-ds > 192.168.3.200.1247: S 2795532295:2795 532295(0) ack 4208572195 win 61680 <mss 1460,nop,wscale 2> (DF) 19:00:15.357911 192.168.3.200.1247 > 192.168.2.7.microsoft-ds: . ack 1 win 52560 (DF) 19:00:18.356736 192.168.3.200.1247 > 192.168.2.7.microsoft-ds: P 1:138(137) ack 1 win 52560 (DF) 19:00:18.357465 192.168.2.7.microsoft-ds > 192.168.3.200.1247: P 1:90(89) ack 13 8 win 61646 (DF) 19:00:18.456659 192.168.3.200.1247 > 192.168.2.7.microsoft-ds: . ack 90 win 5256 0 (DF) 19:00:20.611227 192.168.3.200.1247 > 192.168.2.7.microsoft-ds: P 138:336(198) ac k 90 win 52560 (DF) 19:00:20.612037 192.168.2.7.microsoft-ds > 192.168.3.200.1247: P 90:359(269) ack 336 win 61597 (DF) I see two problems here (in fact its only one problem with a side effect): 1) When sending the ccnew option T/TCP send data with it. 2) As a results the sender times out and retransmit at 19:00:18.356736 adding 3 seconds to the connection data transfer. Now, in the rfc1644 it is stipulated that a host can send data with the cc.new option. I am a bit confused here since T/TCP should accelerate data transfer for small transaction (saving 1 RTT) and not timing out if the destination host does not support it. I think we should avoid sending data with the cc.new option. Any thoughs? Karim Fodil-Lemelin Network Engineering. Xiphos Technologies www.xiphos.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?JCEDLMKGMLLELHAJNIHJGEGHEJAA.kfl>