From mboxrd@z Thu Jan 1 00:00:00 1970 From: Larkin Lowrey Subject: Re: Raid5 device hangs in active state Date: Sun, 11 Mar 2012 19:18:36 -0500 Message-ID: <4F5D40DC.9080304@nuclearwinter.com> References: <4F0A129E.5020706@nuclearwinter.com> <4F5D3562.6060507@shiftmail.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <4F5D3562.6060507@shiftmail.org> Sender: linux-raid-owner@vger.kernel.org To: Asdo Cc: linux-raid List-Id: linux-raid.ids Observation: Setting vm/dirty_background_bytes to 1MB increased my reshape speed by 50% and reduced the resize2fs write interval from once per 5-6 seconds to once per 3-4 seconds. Changing vm/dirty_bytes didn't have any noticeable affect until it got below 50MB at which point performance dropped dramatically. Prior to dialing the knobs, dirty_ratio was 20% and dirty_background_ratio was 10%. The system has 8GB installed. On 3/11/2012 6:29 PM, Asdo wrote: > On 01/08/12 23:03, Larkin Lowrey wrote: >> ... >> Interestingly, to me, when in this state, 'iostat -dx 1' shows the root >> ... > > Do you have any "avgqu-sz" stuck at nonzero value in iostat -dx 1? Yes, avgqu-sz is 1.00. > > > BTW this bug could be related to the recent post by Milan Broz, > subject: "md raid5 fsync deadlock" . > Because also in your case fsyncs are not likely to complete... correct? I do get writes, now about one 4k block every 3-4 seconds. It is quite possible that our two issues are related. --Larkin