From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Thu, 18 Jan 2001 20:46:34 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Thu, 18 Jan 2001 20:44:28 -0500 Received: from ns.virtualhost.dk ([195.184.98.160]:4111 "EHLO virtualhost.dk") by vger.kernel.org with ESMTP id ; Thu, 18 Jan 2001 20:44:05 -0500 Date: Fri, 19 Jan 2001 02:43:51 +0100 From: Jens Axboe To: Marcelo Tosatti Cc: Andrea Arcangeli , linux-kernel@vger.kernel.org Subject: Re: 2.4.1pre8 slowdown on dbench tests Message-ID: <20010119024351.C18209@suse.de> In-Reply-To: <20010119011629.C32087@athlon.random> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: ; from marcelo@conectiva.com.br on Thu, Jan 18, 2001 at 09:51:03PM -0200 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Jan 18 2001, Marcelo Tosatti wrote: > > Marcelo can you give a try with `high_queued_sectors = total_ram / 3' and > > low_queued_sectors = high_queued_sectors / 2 and drop the big ram machine > > check? > > Andrea, > > With the changes you suggested I got almost the same results with > pre8. Good, so it's getting closer. Actually, the dbench numbers being this close (or better) is only due to blk-xx optimizations. The latency is much smaller, which usually really hurts dbench quite a bit in testing. -- * Jens Axboe * SuSE Labs - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/