From owner-freebsd-current@FreeBSD.ORG Thu Feb 10 00:02:06 2005 Return-Path: Delivered-To: freebsd-current@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id B7AEA16A4DD; Thu, 10 Feb 2005 00:02:06 +0000 (GMT) Received: from cyrus.watson.org (cyrus.watson.org [204.156.12.53]) by mx1.FreeBSD.org (Postfix) with ESMTP id 874E843D31; Thu, 10 Feb 2005 00:02:06 +0000 (GMT) (envelope-from robert@fledge.watson.org) Received: from fledge.watson.org (fledge.watson.org [204.156.12.50]) by cyrus.watson.org (Postfix) with SMTP id 2ACF146B45; Wed, 9 Feb 2005 19:02:06 -0500 (EST) Date: Thu, 10 Feb 2005 00:01:01 +0000 (GMT) From: Robert Watson X-Sender: robert@fledge.watson.org To: Scott Long In-Reply-To: <420AA08C.8090809@freebsd.org> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII cc: freebsd-current@freebsd.org Subject: Re: MySQL benchmarks X-BeenThere: freebsd-current@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list List-Id: Discussions about the use of FreeBSD-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 10 Feb 2005 00:02:06 -0000 On Wed, 9 Feb 2005, Scott Long wrote: > > The scalability results look promising. Also, has anyone seen what > > effect WITH_PROC_SCOPE_PTH has on performance? > > It's a mixed bag that has been in significant flux over the past 8 > months due to threading and scheduling infrastructure sometimes fixing > bugs and sometimes introducing new bugs. On the mysql SuperSmack test, > the best results I found on a dual 3GHz Xeon were with system scope > threads under libpthread. libthr was a close second (though all threads > there are process scope by definition) and process scope libpthread was > almost as bad as libc_r. However, that was back in August, and I think > that much has changed since then. Other, non-mysql tests that I've run > recently have shown that process scope libpthread is now the overall > winner. It would be nice to come up with a new matrix of results based > on scheduler, preemption, thread library, and thread attributes. Now if > only I had the 2 days free to do that... David Xu's recent work on threading also looks very promising, and in his benchmarks seemed to substantially outperform MySQL running linuxthreads, libkse, libthr, and libc_r on FreeBSD. I recently set him up with a dual Xeon box to use in the netperf cluster as he was previously benchmarking only on PIII hardware. Robert N M Watson