From mboxrd@z Thu Jan 1 00:00:00 1970 From: Bill Davidsen Subject: Re: Data-check brings system to a standstill Date: Tue, 22 Jun 2010 14:50:38 -0400 Message-ID: <4C2105FE.8020201@tmr.com> References: <4C0FCFCD.4070201@quo.to> <4C19015B.3020807@tmr.com> <4C2036CB.7090505@quo.to> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <4C2036CB.7090505@quo.to> Sender: linux-raid-owner@vger.kernel.org To: Jordan Russell Cc: linux-raid@vger.kernel.org List-Id: linux-raid.ids Jordan Russell wrote: > On 6/16/2010 11:52 AM, Bill Davidsen wrote: > >> Alternatively, you can try setting your io scheduler to >> deadline, >> > > Wow - huge difference! No "hung task" warnings, no processes stuck in > "D" state; the system remains completely responsive throughout the > entire check. > > I don't see why, though, processes are blocking for 2+ minutes on the > simplest of I/O operations when the default CFQ scheduler is used. Isn't > CFQ supposed to distribute I/O bandwidth in a "completely fair" manner? > Do md checks bypass the fairness algorithm? > > The CFQ has its own idea of what is "fair" and works poorly (even abysmally as you saw) for some load cases. -- Bill Davidsen "We can't solve today's problems by using the same thinking we used in creating them." - Einstein