From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Dan A. Dickey" Subject: Re: TCP Hang (or close to it): Windows XP -> Linux (w/tcp_bic) Date: Fri, 29 Oct 2004 13:21:25 -0500 Sender: netdev-bounce@oss.sgi.com Message-ID: <200410291321.25723.dan.dickey@savvis.net> References: <200410290925.48955.dan.dickey@savvis.net> <20041029110019.4f175979.davem@davemloft.net> Reply-To: dan.dickey@savvis.net Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Cc: Sridhar Samudrala , netdev@oss.sgi.com Return-path: To: "David S. Miller" In-Reply-To: <20041029110019.4f175979.davem@davemloft.net> Content-Disposition: inline Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org On Friday 29 October 2004 13:00, David S. Miller wrote: > On Fri, 29 Oct 2004 10:10:42 -0700 (PDT) > > Sridhar Samudrala wrote: > > This looks like the same TCP window overflow problem in 2.6.7 > > that came up a couple of days back. > > > > See the thread > > http://marc.theaimsgroup.com/?t=109887532400001&r=1&w=2 > > and the patch posted by davem. > > Yes, and BTW for the original poster, BIC only effects sender > behavior whereas in your test case the Linux system with BIC > enabled is the receiver. That's a good point. I was concerned that it might be playing a role in what window size got sent back to the sending machine. Glad to hear it's not a worry. BTW - any idea if tcp_bic would be easily modified to be configurable on a per interface basis? We typically have a 100Mb interface on a LAN; and another 100Mb interface going out to a WAN (DS3 or E3). We're using tcp_bic to get near 95+% utilization of the WAN connection; but I don't think we need it on the LAN, and in some situations it may interfere... or, is this not a concern? Thanks again all. Sridhar, thank's for the pointer. I'll be updating to 2.6.9 and will let you know the results of my testing. -Dan -- Dan A. Dickey dan.dickey@savvis.net SAVVIS Transforming Information Technology