From mboxrd@z Thu Jan 1 00:00:00 1970 From: Vladimir Davydov Subject: Re: [patch] mm: memcontrol: revert use of root_mem_cgroup res_counter Date: Fri, 5 Sep 2014 17:55:07 +0400 Message-ID: <20140905135507.GE25641@esperanza> References: <1409921037-21405-1-git-send-email-hannes@cmpxchg.org> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Return-path: Content-Disposition: inline In-Reply-To: <1409921037-21405-1-git-send-email-hannes-druUgvl0LCNAfugRpC6u6w@public.gmane.org> Sender: cgroups-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-ID: Content-Transfer-Encoding: 7bit To: Johannes Weiner Cc: Linus Torvalds , Dave Hansen , Michal Hocko , Andrew Morton , Hugh Dickins , Tejun Heo , linux-mm-Bw31MaZKKs3YtjvyW6yDsg@public.gmane.org, cgroups-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org On Fri, Sep 05, 2014 at 08:43:57AM -0400, Johannes Weiner wrote: > Dave Hansen reports a massive scalability regression in an uncontained > page fault benchmark with more than 30 concurrent threads, which he > bisected down to 05b843012335 ("mm: memcontrol: use root_mem_cgroup > res_counter") and pin-pointed on res_counter spinlock contention. > > That change relied on the per-cpu charge caches to mostly swallow the > res_counter costs, but it's apparent that the caches don't scale yet. > > Revert memcg back to bypassing res_counters on the root level in order > to restore performance for uncontained workloads. > > Reported-by: Dave Hansen > Signed-off-by: Johannes Weiner > Tested-by: Dave Hansen > Acked-by: Michal Hocko It's a pity we have to revert this nice cleanup, but seems we can't do anything better right now. FWIW, Reviewed-by: Vladimir Davydov