From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jens Axboe Subject: Re: [patch] md: pass down BIO_RW_SYNC in raid{1,10} Date: Tue, 9 Jan 2007 08:12:25 +0100 Message-ID: <20070109071224.GG11203@kernel.dk> References: <20070108090833.GF8125@soda.linbit> <20070108150242.1b39eda2.akpm@osdl.org> <17826.55504.264610.856196@notabene.brown> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <17826.55504.264610.856196@notabene.brown> Sender: linux-raid-owner@vger.kernel.org To: Neil Brown Cc: Andrew Morton , Lars Ellenberg , Ingo Molnar , linux-raid@vger.kernel.org, stable@kernel.org List-Id: linux-raid.ids On Tue, Jan 09 2007, Neil Brown wrote: > On Monday January 8, akpm@osdl.org wrote: > > On Mon, 8 Jan 2007 10:08:34 +0100 > > Lars Ellenberg wrote: > > > > > md raidX make_request functions strip off the BIO_RW_SYNC flag, > > > thus introducing additional latency. > > > > > > fixing this in raid1 and raid10 seems to be straight forward enough. > > > > > > for our particular usage case in DRBD, passing this flag improved > > > some initialization time from ~5 minutes to ~5 seconds. > > > > That sounds like a significant fix. > > > > This patch also applies to 2.6.19 and I have tagged it for a -stable > > backport. Neil, are you OK with that? > > Yes, I'm OK with that, thanks. Ack from me as well, it's really a quite nasty bug from a performance POV. Not just for DRDB, but for io schedulers as well. -- Jens Axboe