From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sha Zhengju Subject: Re: [PATCH 3/4] memcg: add per cgroup writeback pages accounting Date: Sat, 24 Aug 2013 00:11:50 +0800 Message-ID: References: <1377165190-24143-1-git-send-email-handai.szj@taobao.com> <20130822154002.ce4310d865ede3a0d30f0ce8@linux-foundation.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Cc: Michal Hocko , KAMEZAWA Hiroyuki , Greg Thelen , "linux-mm@kvack.org" , Cgroups , "linux-fsdevel@vger.kernel.org" , Sha Zhengju To: Andrew Morton Return-path: Received: from mail-bk0-f47.google.com ([209.85.214.47]:36248 "EHLO mail-bk0-f47.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753941Ab3HWQLw (ORCPT ); Fri, 23 Aug 2013 12:11:52 -0400 In-Reply-To: <20130822154002.ce4310d865ede3a0d30f0ce8@linux-foundation.org> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Fri, Aug 23, 2013 at 6:40 AM, Andrew Morton wrote: > On Thu, 22 Aug 2013 17:53:10 +0800 Sha Zhengju wrote: > >> This patch is to add memcg routines to count writeback pages > > Well OK, but why? What use is the feature? In what ways are people > suffering due to its absence? My apologies for not explaining it clearly. It's subset of memcg dirty page accounting(including dirty, writeback, nfs_unstable pages from a broad sense), which can provide a more sound knowledge of memcg behavior. That would be straightforward to add new features like memcg dirty page throttling and even memcg aware flushing. However, the dirty one is more complicated and performance senstive, so I need more efforts to improve it and let the writeback patch go first. Afterwards I'll only focus on dirty page itself. -- Thanks, Sha