From: "Alexander V. Lukyanov" Subject: Re: [PATCH] ext4: allow inode_readahead_blks=0 (linux-2.6.37) Date: Thu, 24 Feb 2011 11:18:06 +0300 Message-ID: <20110224081806.GA14787@lw.yar.ru> References: <20110208063925.GA13619@lw.yar.ru> <20110222023211.GF2924@thunk.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii To: "Ted Ts'o" , linux-ext4@vger.kernel.org, Andreas Dilger , linux-kernel@vger.kernel.org Return-path: Received: from lav-w.yar.ru ([193.233.48.101]:55694 "EHLO localhost.localdomain" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1755357Ab1BXIS3 (ORCPT ); Thu, 24 Feb 2011 03:18:29 -0500 Content-Disposition: inline In-Reply-To: <20110222023211.GF2924@thunk.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Mon, Feb 21, 2011 at 09:32:11PM -0500, Ted Ts'o wrote: > On Tue, Feb 08, 2011 at 09:39:25AM +0300, Alexander V. Lukyanov wrote: > > Hello! > > > > I cannot disable inode-read-ahead feature of ext4 (on 2.6.37): > > > > # echo 0 > /sys/fs/ext4/sda2/inode_readahead_blks > > bash: echo: write error: Invalid argument > > > > On a server with lots of small files and random access this read-ahead makes > > performance worse, and I'd like to disable it. I work around this problem > > by using value of 1, but it still reads an extra block. > > So I'm curious --- have you actually benchmarked a performance > decrease? What sort of hardware are you using? Yes, with the default value of inode_readahead_blks LA went from 4 to 30 (if I remember correctly). The problem was the increased load on HDD. The hardware is: Core2duo CPU, 4GB RAM, 4x80GB SATA disks without NCQ, the load is evenly distributed on the disks. At that time each disk contained 1 million files, randomly accessed for read/create-write, 10MB/s read and 10MB/s write (rate sum of 4 disks). > The readahead should be changing a 4k read to a 8k read with a value > of 1, which shouldn't take a much of a difference to a HDD. Sure, with inode_readahead_blks=1 it works acceptably. But I'd like to disable the inode read-ahead completely. > I can apply this patch, but is it really making a difference for you? I think it is logical to be able to disable an unneeded feature. Besides, there is a code already to check s_inode_readahead_blks!=0 (fs/ext4/inode.c:4737): /* * If we need to do any I/O, try to pre-readahead extra * blocks from the inode table. */ if (EXT4_SB(sb)->s_inode_readahead_blks) { -- Alexander.