From owner-freebsd-fs@FreeBSD.ORG Fri Oct 14 16:33:32 2005 Return-Path: X-Original-To: fs@freebsd.org Delivered-To: freebsd-fs@FreeBSD.ORG Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id A041D16A41F for ; Fri, 14 Oct 2005 16:33:32 +0000 (GMT) (envelope-from rick@snowhite.cis.uoguelph.ca) Received: from mailhub.cs.uoguelph.ca (mailhub.cs.uoguelph.ca [131.104.96.75]) by mx1.FreeBSD.org (Postfix) with ESMTP id 4D7AE43D45 for ; Fri, 14 Oct 2005 16:33:32 +0000 (GMT) (envelope-from rick@snowhite.cis.uoguelph.ca) Received: from snowhite.cis.uoguelph.ca (snowhite.cis.uoguelph.ca [131.104.48.1]) by mailhub.cs.uoguelph.ca (8.13.1/8.13.1) with ESMTP id j9EGXViE009821 for ; Fri, 14 Oct 2005 12:33:31 -0400 Received: (from rick@localhost) by snowhite.cis.uoguelph.ca (8.9.3/8.9.3) id MAA22435 for fs@freebsd.org; Fri, 14 Oct 2005 12:34:55 -0400 (EDT) Date: Fri, 14 Oct 2005 12:34:55 -0400 (EDT) From: rick@snowhite.cis.uoguelph.ca Message-Id: <200510141634.MAA22435@snowhite.cis.uoguelph.ca> To: fs@freebsd.org X-Scanned-By: MIMEDefang 2.52 on 131.104.96.75 Cc: Subject: FreeBSD NFS server not responding to TCP SYN packets from Linux/SunOS clients X-BeenThere: freebsd-fs@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Filesystems List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 14 Oct 2005 16:33:32 -0000 > ps: It would be nice if someone with the right expertise could explore > other things in TCP specifically for NFS. For example, I don't see > why a retransmit timeout should go above about 100msec > > nfs/rpc shouldn't retransmit at all over tcp except when there has been a > reconnect. Tcp might retransmit, but modern implementations will always > choose the right timeout dynamically, unless packet loss is excessive. > Yes, I was referring to TCP retransmits and I was suggesting that TCP might not be "choosing the right timeout dynamically". A lot of what is in TCP timers now dates back to work done w.r.t. congestion avoidance (very good work, I might add) in the late 1980s. I don't know if anyone has revisited this, now that almost everything is gigibit fibre (in the late 1980s, there were still 56Kbps internet links out there). I don't know enough about TCP (and haven't looked at recent literature), but I do know that I see retransmit timeouts back off to 1-2sec quite rapidly, as soon as I introduce a 20msec transit delay between client<->server and throw away the odd packet. The result is abismal NFS/TCP preformance. rick