Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755465AbaAHImP (ORCPT ); Wed, 8 Jan 2014 03:42:15 -0500 Received: from e28smtp08.in.ibm.com ([122.248.162.8]:37430 "EHLO e28smtp08.in.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755198AbaAHImC (ORCPT ); Wed, 8 Jan 2014 03:42:02 -0500 Message-ID: <52CD1113.2070003@linux.vnet.ibm.com> Date: Wed, 08 Jan 2014 14:19:23 +0530 From: Raghavendra K T Organization: IBM User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130625 Thunderbird/17.0.7 MIME-Version: 1.0 To: Andrew Morton CC: Fengguang Wu , David Cohen , Al Viro , Damien Ramonda , jack@suse.cz, Linus , linux-mm@kvack.org, linux-kernel@vger.kernel.org Subject: Re: [RFC PATCH V3] mm readahead: Fix the readahead fail in case of empty numa node References: <1389003715-29733-1-git-send-email-raghavendra.kt@linux.vnet.ibm.com> <20140106141300.4e1c950d45c614d6c29bdd8f@linux-foundation.org> In-Reply-To: <20140106141300.4e1c950d45c614d6c29bdd8f@linux-foundation.org> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit X-TM-AS-MML: disable X-Content-Scanned: Fidelis XPS MAILER x-cbid: 14010808-2000-0000-0000-00000F382BE0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 01/07/2014 03:43 AM, Andrew Morton wrote: > On Mon, 6 Jan 2014 15:51:55 +0530 Raghavendra K T wrote: > >> + /* >> + * Readahead onto remote memory is better than no readahead when local >> + * numa node does not have memory. We sanitize readahead size depending >> + * on free memory in the local node but limiting to 4k pages. >> + */ >> + return local_free_page ? min(sane_nr, local_free_page / 2) : sane_nr; >> } > > So if the local node has two free pages, we do just one page of > readahead. > > Then the local node has one free page and we do zero pages readahead. > > Assuming that bug(!) is fixed, the local node now has zero free pages > and we suddenly resume doing large readahead. > > This transition from large readahead to very small readahead then back > to large readahead is illogical, surely? > > Hi Andrew, Thanks for having a look at this. You are correct that there is a transition from small readahead to large once we have zero free pages. I am not sure I can defend well, but 'll give a try :). Hoping that we have evenly distributed cpu/memory load, if we have very less free+inactive memory may be we are in really bad shape already. But in the case where we have a situation like below [1] (cpu does not have any local memory node populated) I had mentioned earlier where we will have to depend on remote node always, is it not that sanitized readahead onto remote memory seems better? But having said that I am not able to get an idea of sane implementation to solve this readahead failure bug overcoming the anomaly you pointed :(. hints/ideas.. ?? please let me know. [1]: IBM P730 ---------------------------------- # numactl -H available: 2 nodes (0-1) node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 node 0 size: 0 MB node 0 free: 0 MB node 1 cpus: node 1 size: 12288 MB node 1 free: 10440 MB node distances: node 0 1 0: 10 40 1: 40 10 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/