From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1754501Ab2AaANH (ORCPT <rfc822;w@1wt.eu>);
	Mon, 30 Jan 2012 19:13:07 -0500
Received: from mga03.intel.com ([143.182.124.21]:29604 "EHLO mga03.intel.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1752838Ab2AaANG (ORCPT <rfc822;linux-kernel@vger.kernel.org>);
	Mon, 30 Jan 2012 19:13:06 -0500
X-ExtLoop1: 1
X-IronPort-AV: E=Sophos;i="4.71,315,1320652800"; 
   d="scan'208";a="101328457"
Date: Mon, 30 Jan 2012 22:48:21 +0800
From: Wu Fengguang <wfg@linux.intel.com>
To: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Herbert Poetzl <herbert@13thfloor.at>,
        Andrew Morton <akpm@linux-foundation.org>,
        LKML <linux-kernel@vger.kernel.org>, Jens Axboe <axboe@kernel.dk>,
        Tejun Heo <tj@kernel.org>, Li Shaohua <shaohua.li@intel.com>
Subject: Re: Bad SSD performance with recent kernels
Message-ID: <20120130144821.GA23671@localhost>
References: <20120127060034.GG29272@MAIL.13thfloor.at>
 <20120128125108.GA9661@localhost>
 <1327757611.7199.6.camel@edumazet-laptop>
 <20120129055917.GB8513@localhost>
 <1327831380.14602.6.camel@edumazet-laptop>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <1327831380.14602.6.camel@edumazet-laptop>
User-Agent: Mutt/1.5.20 (2009-06-14)
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Sun, Jan 29, 2012 at 11:03:00AM +0100, Eric Dumazet wrote:

> Very strange, my bissection ended on following commit :
> 
> commit 805f6b5e1cbfedfb9b3d354013e7f4b13a79270f
> Author: Tao Ma <boyu.mt@taobao.com>
> Date:   Fri Mar 11 20:11:59 2011 +0100
> 
>     blktrace: Use rq->cmd_flags directly in blk_add_trace_rq.
> 
> This makes no sense.

Yeah, commit 55602dd66f535 ("fs: make generic file read/write
functions plug") looks like the real root cause and should probably
be reverted as a whole. Buffered read/write syscalls are not directly
tied to IO after all (readahead/writeback have the right plug points).

Thanks,
Fengguang

> hdparm uses 2MB block reads, so read_ahead (128KB) is too small for best
> perf
> 
> # cat /sys/class/block/sda/queue/read_ahead_kb
> 128
> 
> # dd if=/dev/sda of=/dev/null bs=128k 
> ^C
> 63744+0 enregistrements lus
> 63743+0 enregistrements écrits
> 8354922496 octets (8,4 GB) copiés, 39,975 s, 209 MB/s
> 
> # hdparm -t /dev/sda
> 
> /dev/sda:
>  Timing buffered disk reads: 510 MB in  3.00 seconds = 169.75 MB/sec
> 
> # uname -a
> Linux edumazet-laptop 3.0.0-15-generic #26-Ubuntu SMP Fri Jan 20
> 17:23:00 UTC 2012 x86_64 x86_64 x86_64 GNU/Linux
> 
>