From owner-freebsd-hackers@FreeBSD.ORG Sat Jan 31 21:11:44 2004 Return-Path: Delivered-To: freebsd-hackers@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id C265F16A4CE for ; Sat, 31 Jan 2004 21:11:44 -0800 (PST) Received: from server.lab.cyberlifelabs.com (adsl-64-142-24-131.sonic.net [64.142.24.131]) by mx1.FreeBSD.org (Postfix) with ESMTP id A830243D1F for ; Sat, 31 Jan 2004 21:11:43 -0800 (PST) (envelope-from milo@cyberlifelabs.com) Received: from [192.168.1.3] (beastie.lab.cyberlifelabs.com [192.168.1.3]) i115Bhvn068145 for ; Sat, 31 Jan 2004 21:11:43 -0800 (PST) (envelope-from milo@cyberlifelabs.com) From: Milo Hyson To: freebsd-hackers@freebsd.org In-Reply-To: <1075600839.50745.7.camel@beastie.lab.cyberlifelabs.com> References: <1075600839.50745.7.camel@beastie.lab.cyberlifelabs.com> Content-Type: text/plain Message-Id: <1075612303.340.2.camel@beastie.lab.cyberlifelabs.com> Mime-Version: 1.0 X-Mailer: Ximian Evolution 1.4.5 Date: Sat, 31 Jan 2004 21:11:43 -0800 Content-Transfer-Encoding: 7bit Subject: Re: Weird context-switching performance [RESOLVED] X-BeenThere: freebsd-hackers@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list List-Id: Technical Discussions relating to FreeBSD List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sun, 01 Feb 2004 05:11:44 -0000 Situation resolved. Turns out there was one small line in a kernel config file that we overlooked. Seems the apm device was causing a massive performance penalty. We don't really need it so we removed it. All performance metrics are now exactly where they should be. -- Milo Hyson Chief "Mad" Scientist CyberLife Labs, LLC On Sat, 2004-01-31 at 18:00, Milo Hyson wrote: > We don't have any other systems that are as similar as the two in the > test, however we did plot context-switching performance against CPU and > memory performance on several systems to see if anything jumped out. We > found that beastie (the 2200+) is doing only about 26% of the > task-switches/dhrystone that the other systems were, which were all > pretty much about equal with each other. This is consistent with the > observation that appserver (the 2100+) is four-times as fast. I think we > can probably conclude that beastie is running slower than it should. > > So the question still remains. What could affect context-switching to > this degree yet not show up in other benchmarks?