From: Alexey Lyahkov Subject: Re: page eviction from the buddy cache Date: Mon, 22 Apr 2013 15:18:59 +0300 Message-ID: <51874517-8D82-433E-9E10-13167C736D5F@gmail.com> References: <51504A40.6020604@ya.ru> <20130327150743.GC14900@thunk.org> <3C8EEEF8-C1EB-4E3D-8DE6-198AB1BEA8C0@gmail.com> <515CD665.9000300@gmail.com> <239AD30A-2A31-4346-A4C7-8A6EB8247990@gmail.com> <51730619.3030204@fastmail.fm> Mime-Version: 1.0 (Apple Message framework v1283) Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 8BIT Cc: Will Huck , Hugh Dickins , Theodore Ts'o , Andrew Perepechko , linux-ext4@vger.kernel.org, akpm@linux-foundation.org, linux-mm@kvack.org To: Bernd Schubert Return-path: Received: from mail-we0-f175.google.com ([74.125.82.175]:62695 "EHLO mail-we0-f175.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751304Ab3DVMTL convert rfc822-to-8bit (ORCPT ); Mon, 22 Apr 2013 08:19:11 -0400 Received: by mail-we0-f175.google.com with SMTP id t11so6381335wey.34 for ; Mon, 22 Apr 2013 05:19:10 -0700 (PDT) In-Reply-To: <51730619.3030204@fastmail.fm> Sender: linux-ext4-owner@vger.kernel.org List-ID: Bernd, I fact marking REQ_META will help with io scheduler(). I and Andrew will discuss it's some time ago, but don't expect a too much performance improvements anyway - i will pickup patch from Intel jira and ask performance engineer to evaluate it. bw. My problem is in memory caching. I may rewrite a buddy cache to avoid using a page cache at all as these pages but LRU aging will lost. On Apr 21, 2013, at 00:18, Bernd Schubert wrote: > Alex, Andrew, > > did you notice the patch Ted just sent? > ("ext4: mark all metadata I/O with REQ_META") > > I would like to see a way to mark pages read in with REQ_META to be kept > in cache preferred over other pages. I guess that would solve LU-15 > (https://jira.hpdd.intel.com/browse/LU-15) and also the direntry-block > issue I tried to solve about 2 years ago > (http://patchwork.ozlabs.org/patch/101200/). But using REQ_META to tag > pages would probably also solve the same issue for other file systems. > Is there anything already in the mm layer that could be used for that? > > Thanks, > Bernd >