From: "Alexander V. Lukyanov" Subject: Re: PATCH: allow inode_readahead_blks=0 Date: Wed, 17 Nov 2010 10:48:54 +0300 Message-ID: <20101117074850.GA29191@proxy> References: <20101116080546.GA7949@proxy> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii To: linux-ext4@vger.kernel.org Return-path: Received: from fire-ce0.netis.ru ([193.233.48.99]:59311 "EHLO fire.netis.ru" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758063Ab0KQHtY (ORCPT ); Wed, 17 Nov 2010 02:49:24 -0500 Received: from proxy.yar.ru (proxy.yar.ru [193.233.48.78]) by fire.netis.ru (8.14.3/8.14.2) with ESMTP id oAH7muFh004113 for ; Wed, 17 Nov 2010 10:48:58 +0300 (MSK) Received: (from root@localhost) by proxy.yar.ru (8.14.4/8.14.2/Submit) id oAH7msHR029419 for linux-ext4@vger.kernel.org; Wed, 17 Nov 2010 10:48:54 +0300 Content-Disposition: inline In-Reply-To: <20101116080546.GA7949@proxy> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Tue, Nov 16, 2010 at 11:05:46AM +0300, Alexander V. Lukyanov wrote: > Currently the value 0 is disallowed, as it is not a power of 2. BTW, the default value inode_readahead_blks=32 gives me troubles: it raises load average and increases request latency on a busy squid server. I explain it by random access to large number of files which are present in the squid cache. BTW2: the value of 0 should be allowed, it is explicitly used by the code: /* * 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.