[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20091109070737.GE3042@balbir.in.ibm.com>
Date: Mon, 9 Nov 2009 12:37:37 +0530
From: Balbir Singh <balbir@...ux.vnet.ibm.com>
To: KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
Cc: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"linux-mm@...ck.org" <linux-mm@...ck.org>,
"nishimura@....nes.nec.co.jp" <nishimura@....nes.nec.co.jp>,
cl@...ux-foundation.org,
"akpm@...ux-foundation.org" <akpm@...ux-foundation.org>
Subject: Re: [PATCH 2/2] memcg : rewrite percpu countings with new
interfaces
* KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com> [2009-11-06 17:55:45]:
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
>
> Now, alloc_percpu() alloc good dynamic allocations and
> Recent updates on percpu.h gives us following kind of ops
> - __this_cpu_add() etc...
> This is designed to be a help for reduce code size in hot-path
> and very useful to handle percpu area. Thanks for great works.
>
> This patch rewrite memcg's (not-good) percpu status with new
> percpu support macros. This decreases code size and instruction
> size. By this, this area is now NUMA-aware and may have performance
> benefit.
>
> I got good result in parallel pagefault test. (my host is 8cpu/2socket)
>
> before==
> Performance counter stats for './runpause.sh' (5 runs):
>
> 474070.055912 task-clock-msecs # 7.881 CPUs ( +- 0.013% )
> 35829310 page-faults # 0.076 M/sec ( +- 0.217% )
> 3803016722 cache-references # 8.022 M/sec ( +- 0.215% ) (scaled from 100.00%)
> 1104083123 cache-misses # 2.329 M/sec ( +- 0.961% ) (scaled from 100.00%)
>
> 60.154982314 seconds time elapsed ( +- 0.018% )
>
> after==
> Performance counter stats for './runpause.sh' (5 runs):
>
> 474919.429670 task-clock-msecs # 7.896 CPUs ( +- 0.013% )
> 36520440 page-faults # 0.077 M/sec ( +- 1.854% )
> 3109834751 cache-references # 6.548 M/sec ( +- 0.276% )
> 1053275160 cache-misses # 2.218 M/sec ( +- 0.036% )
>
> 60.146585280 seconds time elapsed ( +- 0.019% )
>
> This test is affected by cpu-utilization but I think more improvements
> will be found in bigger system.
>
Hi, Kamezawa-San,
Could you please post the IPC results as well?
--
Balbir
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists