From mboxrd@z Thu Jan 1 00:00:00 1970 From: Kamezawa Hiroyuki Subject: Re: [PATCH] memcg: implement boost mode Date: Mon, 01 Apr 2013 19:02:36 +0900 Message-ID: <51595B3C.5090900@jp.fujitsu.com> References: <1364801670-10241-1-git-send-email-glommer@parallels.com> <51595311.7070509@jp.fujitsu.com> <515953AE.3000403@parallels.com> <20130401093740.GA30749@dhcp22.suse.cz> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20130401093740.GA30749-2MMpYkNvuYDjFM9bn6wA6Q@public.gmane.org> Sender: cgroups-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-ID: Content-Type: text/plain; charset="us-ascii"; format="flowed" To: Michal Hocko Cc: Glauber Costa , linux-mm-Bw31MaZKKs3YtjvyW6yDsg@public.gmane.org, Johannes Weiner , Andrew Morton , cgroups-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Tejun Heo (2013/04/01 18:37), Michal Hocko wrote: > On Mon 01-04-13 13:30:22, Glauber Costa wrote: >> On 04/01/2013 01:27 PM, Kamezawa Hiroyuki wrote: >>> (2013/04/01 16:34), Glauber Costa wrote: >>>> There are scenarios in which we would like our programs to run faster. >>>> It is a hassle, when they are contained in memcg, that some of its >>>> allocations will fail and start triggering reclaim. This is not good >>>> for the program, that will now be slower. >>>> >>>> This patch implements boost mode for memcg. It exposes a u64 file >>>> "memcg boost". Every time you write anything to it, it will reduce the >>>> counters by ~20 %. Note that we don't want to actually reclaim pages, >>>> which would defeat the very goal of boost mode. We just make the >>>> res_counters able to accomodate more. >>>> >>>> This file is also available in the root cgroup. But with a slightly >>>> different effect. Writing to it will make more memory physically >>>> available so our programs can profit. >>>> >>>> Please ack and apply. >>>> >>> Nack. >>> >>>> Signed-off-by: Glauber Costa >>> >>> Please update limit temporary. If you need call-shrink-explicitly-by-user, >>> I think you can add it. >>> >> >> I don't want to shrink memory because that will make applications >> slower. I want them to be faster, so they need to have more memory. >> There is solid research backing up my approach: >> http://www.dilbert.com/fast/2008-05-08/ > > :) > ;) -Kame