From owner-freebsd-performance@FreeBSD.ORG Sun Apr 13 23:55:12 2003 Return-Path: Delivered-To: freebsd-performance@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 855E037B401 for ; Sun, 13 Apr 2003 23:55:12 -0700 (PDT) Received: from relay.pair.com (relay.pair.com [209.68.1.20]) by mx1.FreeBSD.org (Postfix) with SMTP id 4D55943FF2 for ; Sun, 13 Apr 2003 23:55:11 -0700 (PDT) (envelope-from silby@silby.com) Received: (qmail 31351 invoked from network); 14 Apr 2003 06:55:09 -0000 Received: from niwun.pair.com (HELO localhost) (209.68.2.70) by relay.pair.com with SMTP; 14 Apr 2003 06:55:09 -0000 X-pair-Authenticated: 209.68.2.70 From: Mike Silbersack To: Hiten Pandya In-Reply-To: <20030412195711.GA30459@unixdaemons.com> Message-ID: <20030413204451.V93049@odysseus.silby.com> References: <200304101311.h3ADBgjY022790@samson.dc.luth.se> <20030412195711.GA30459@unixdaemons.com> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII cc: Borje Josefsson cc: freebsd-performance@freebsd.org cc: Eric Anderson cc: David Gilbert cc: freebsd-hackers@freebsd.org Subject: Re: tcp_output starving -- is due to mbuf get delay? X-BeenThere: freebsd-performance@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list List-Id: Performance/tuning List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 14 Apr 2003 06:55:12 -0000 Date: Mon, 14 Apr 2003 06:55:12 -0000 X-Original-Date: Sun, 13 Apr 2003 20:51:20 -0500 (CDT) X-List-Received-Date: Mon, 14 Apr 2003 06:55:12 -0000 On Sat, 12 Apr 2003, Hiten Pandya wrote: > > That tcp_quench knocks the window size back to one packet, if I'm not > > mistaken. You might want to put a counter there and see if that's > > happening frequently to you; if so, it might explain some loss of > > performance. > > Maybe something like this: > > Cheers. > > -- Hiten As Jayanth pointed out, ip_output already keeps statistics on when it returns ENOBUFS, so adding this additional variable wouldn't be of much benefit. For reference (from netstat -s): 0 output packets dropped due to no bufs, etc. Although, I think ip_output could be improved. It precalculates the expected # of packets to be sent, and returns ENOBUFS early if its estimate looks bad. However, if it gets to the point where if_output is called, and if_output then fails, no statistic is kept. So, it does look like Hiten's patch could be useful, as well as another counter tracking if_output failures. But I'm too busy to worry about them for now. :) Mike "Silby" Silbersack