From owner-freebsd-hackers@FreeBSD.ORG Thu Nov 13 04:36:35 2003 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 39F3D16A4CE for ; Thu, 13 Nov 2003 04:36:35 -0800 (PST) Received: from heron.mail.pas.earthlink.net (heron.mail.pas.earthlink.net [207.217.120.189]) by mx1.FreeBSD.org (Postfix) with ESMTP id 81B7943FBF for ; Thu, 13 Nov 2003 04:36:34 -0800 (PST) (envelope-from tlambert2@mindspring.com) Received: from user-2ivfmpo.dialup.mindspring.com ([165.247.219.56] helo=mindspring.com) by heron.mail.pas.earthlink.net with asmtp (SSLv3:RC4-MD5:128) (Exim 3.33 #1) id 1AKGi8-0003RY-00; Thu, 13 Nov 2003 04:36:33 -0800 Message-ID: <3FB37807.2887043C@mindspring.com> Date: Thu, 13 Nov 2003 04:24:39 -0800 From: Terry Lambert X-Mailer: Mozilla 4.79 [en] (Win98; U) X-Accept-Language: en MIME-Version: 1.0 To: Daniel Ellard References: <20031112103358.S11644@bowser.eecs.harvard.edu> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-ELNK-Trace: b1a02af9316fbb217a47c185c03b154d40683398e744b8a470459bcf3b8945d5fa834d01351b5c49a2d4e88014a4647c350badd9bab72f9c350badd9bab72f9c cc: freebsd-hackers@freebsd.org Subject: Re: Confused about HyperThreading and Performance 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: Thu, 13 Nov 2003 12:36:35 -0000 Daniel Ellard wrote: > Can someone point me at some non-marketing documentation about > hyperthreading on the latest Intel chips? I'm seeing some strange > performance measurements and I would like to figure out what they > mean. Go out to Intel's web site's "developer" section, and look for "SMT". There is a lot of literature. The reason you are seeing a performance drop is contention for shared resources that the scheduler doesn't know are shared. SPECmark and similar benchmarks tend to get worse numbers on every OS when SMT is enabled, due to contention. The only model which will work is a hierarchical affinity and negaffinity model, and I am not away of an OS that is not also treating the hardware as NUMA which works this way (and none of those run on Intel chips; mostly, they run on real NUMA hardware). -- Terry