From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S264647AbTFQK6q (ORCPT ); Tue, 17 Jun 2003 06:58:46 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S264649AbTFQK6q (ORCPT ); Tue, 17 Jun 2003 06:58:46 -0400 Received: from grouse.mail.pas.earthlink.net ([207.217.120.116]:58872 "EHLO grouse.mail.pas.earthlink.net") by vger.kernel.org with ESMTP id S264647AbTFQK6n (ORCPT ); Tue, 17 Jun 2003 06:58:43 -0400 Date: Tue, 17 Jun 2003 07:14:00 -0400 To: piggin@cyberone.com.au, akpm@digeo.com Cc: linux-kernel@vger.kernel.org Subject: Re: i/o benchmarks on 2.5.70* kernels Message-ID: <20030617111400.GA11839@rushmore> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4.1i From: rwhron@earthlink.net Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org > tiobench on SMP results are not very good, lots of > fragmentation, the random IO throughput drops is > probably due to AS strangling TCQ though. You are > using SMP and TCQ, right? SMP yes. The non-benchmarked disks (which don't matter say this at boot time: scsi0:A:1:0: Tagged Queuing enabled. Depth 4 (aic7xxx) These config options on the non-benchmarked disks: CONFIG_AIC7XXX_CMDS_PER_DEVICE=4 CONFIG_AIC7XXX_RESET_DELAY_MS=15000 CONFIG_AIC7XXX_DEBUG_ENABLE=y CONFIG_AIC7XXX_DEBUG_MASK=0 CONFIG_AIC7XXX_REG_PRETTY_PRINT=y ---- The benchmarked disks are on scsi3 and scsi4. The benchmarked disks do not tell what TCQ they have. CONFIG_SCSI_QLOGIC_ISP=y CONFIG_SCSI_QLOGIC_FC=y CONFIG_SCSI_QLOGIC_FC_FIRMWARE=y TCQ on the benchmarked disks may be 1024... +#define MAXISPREQUEST(isp) ((IS_FC(isp) || IS_ULTRA2(isp))? 1024 : 256) but that's just a guess. -- Randy Hron http://home.earthlink.net/~rwhron/kernel/bigbox.html