From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Wed, 26 Mar 2003 10:21:02 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Wed, 26 Mar 2003 10:21:02 -0500 Received: from franka.aracnet.com ([216.99.193.44]:21376 "EHLO franka.aracnet.com") by vger.kernel.org with ESMTP id ; Wed, 26 Mar 2003 10:21:00 -0500 Date: Wed, 26 Mar 2003 07:32:09 -0800 From: "Martin J. Bligh" To: linux-kernel Subject: [Bug 503] New: Hyper Threading technology Message-ID: <227380000.1048692729@[10.10.2.4]> X-Mailer: Mulberry/2.2.1 (Linux/x86) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Content-Disposition: inline Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org http://bugme.osdl.org/show_bug.cgi?id=503 Summary: Hyper Threading technology Kernel Version: 2.5.65 Status: NEW Severity: low Owner: rml@tech9.net Submitter: koch@dfki.uni-kl.de Distribution: SuSE 8.0 Hardware Environment: Tyan Thunder i860 S2603, dual XEON 2.8GHz, 512MB RDRAM Software Environment: Problem Description: System hung after enabling HT in BIOS Steps to reproduce: I just installed SuSE 8.0 on the machine, downloaded the kernel, configured it (i think to sane settings), put it into /boot and ran lilo. After reboot, i did some benchmarking and the machine seemed to run fine. When i rebooted a second time and enabled HT in the BIOS, the machine came up but hung after i logged in (before i could run any benchmark). I tried everything, but had to use hardware reset. I didn't investigate this further because the machine had to be finished yesterday. And now something completely different: Did i say benchmarking? No, not really. I did compile the 2.5.65 kernel with several different 'make -j' options to see the impact of HT and if i should enable or disable it and what the benefit of using a 2.5 kernel would be. I know that compiling a kernel is not exactly a benchmark, but since the machine will be used in software development under C i thought it might reveal some issues anyway. What i found out is interesting (at least for me) and did not meet my (and other peoples) expectations: The penalty of using the 2.5.65 kernel (4.6%) is bigger than the penalty of enabling HT on the 2.4.18 that came with SuSE 8.0 (only 2%). The gain of using HT on the 2.4.18 is quite astonishing: about 14%. Here are my numbers: 2 x XEON 2.8GHz (HT enabled) SuSE 8.0 Kernel 2.4.18 real user sys make - 243.6 229.0 23.6 make -j 2 - 132.9 247.6 26.2 make -j 3 - 120.8 335.3 30.8 make -j 4 - 115.8 406.1 34.6 make -j 5 - 116.0 406.0 34.8 2 x XEON 2.8GHz (HT disabled) SuSE 8.0 Kernel 2.4.18 make - 239.0 223.8 22.3 make -j 2 - 132.3 229.7 23.4 make -j 3 - 132.8 231.3 24.5 2 x XEON 2.8GHz (HT enabled) SuSE 8.0 Kernel 2.5.65 --------- System hung before any benchmark was run --------- 2 x XEON 2.8GHz (HT disabled) SuSE 8.0 Kernel 2.5.65 make - 250.5 229.9 25.0 make -j 2 - 134.9 235.8 27.1 make -j 3 - 136.7 237.3 27.5