From owner-freebsd-net@FreeBSD.ORG Tue Jan 11 19:14:39 2005 Return-Path: Delivered-To: freebsd-net@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id EB4EA16A4CE for ; Tue, 11 Jan 2005 19:14:39 +0000 (GMT) Received: from sb.santaba.com (sb.santaba.com [207.154.84.26]) by mx1.FreeBSD.org (Postfix) with ESMTP id A46E143D39 for ; Tue, 11 Jan 2005 19:14:39 +0000 (GMT) (envelope-from jbehl@fastclick.com) Received: from [192.168.3.100] (unknown [205.180.85.193]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (No client certificate requested) by sb.santaba.com (Postfix) with ESMTP id 0F29128433; Tue, 11 Jan 2005 11:14:39 -0800 (PST) Message-ID: <41E425BD.709@fastclick.com> Date: Tue, 11 Jan 2005 11:15:09 -0800 From: Jeff Behl User-Agent: Mozilla Thunderbird 0.8 (Windows/20040913) X-Accept-Language: en-us, en MIME-Version: 1.0 To: Mohan Srinivasan References: <20050111185033.73879.qmail@web80602.mail.yahoo.com> In-Reply-To: <20050111185033.73879.qmail@web80602.mail.yahoo.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit cc: freebsd-net@freebsd.org Subject: Re: %cpu in system - squid performance in FreeBSD 5.3 X-BeenThere: freebsd-net@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list List-Id: Networking and TCP/IP with FreeBSD List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 11 Jan 2005 19:14:40 -0000 Yes, I believe the kqueue version of squid would show much better results. Unfortunately it fails to compile and I have yet the time to try mucking with it more. I'll get back to the list when I am able to get it up and running... jeff Mohan Srinivasan wrote: >Following up to a mail from Jeff Behl and Sean Chittenden back in Dec. > >http://lists.freebsd.org/pipermail/freebsd-net/2004-December/006074.html > >From your description, it looks like moving a kqueue based Squid will >help considerably (it looks like there is a version of Squid that >is kqueue based - not sure how stable that is though). If you drop a quick >kernel profile, you will see most of the system CPU being spent in select() >caused polling of descriptors. In my previous experience with a Squid-based >proxy several years ago, once you dropped more than a couple of hundred >connections into select(), CPU utilization spiked sharply because of >the descriptor polling. > >We then hoisted Squid on top of a (homebrew) version of kqueue, which >caused system CPU to drop dramatically, because all the descriptor polling >was avoided. > >mohan > > >