From mboxrd@z Thu Jan 1 00:00:00 1970 From: Brian King Subject: Re: IPR driver performance issue Date: Thu, 25 Aug 2005 15:11:03 +0100 Message-ID: <430DD177.3080208@us.ibm.com> References: <20050825022411.GA6057@localhost.localdomain> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: cmm@us.ibm.com, wmb@us.ibm.com, slpratt@us.ibm.com, greg@kroah.com, linux-fsdevel@vger.kernel.org Return-path: Received: from e6.ny.us.ibm.com ([32.97.182.146]:9923 "EHLO e6.ny.us.ibm.com") by vger.kernel.org with ESMTP id S965010AbVHYOLJ (ORCPT ); Thu, 25 Aug 2005 10:11:09 -0400 Received: from d01relay02.pok.ibm.com (d01relay02.pok.ibm.com [9.56.227.234]) by e6.ny.us.ibm.com (8.12.11/8.12.11) with ESMTP id j7PEB7H1025727 for ; Thu, 25 Aug 2005 10:11:07 -0400 Received: from d01av01.pok.ibm.com (d01av01.pok.ibm.com [9.56.224.215]) by d01relay02.pok.ibm.com (8.12.10/NCO/VERS6.7) with ESMTP id j7PEB7lL261714 for ; Thu, 25 Aug 2005 10:11:07 -0400 Received: from d01av01.pok.ibm.com (loopback [127.0.0.1]) by d01av01.pok.ibm.com (8.12.11/8.13.3) with ESMTP id j7PEB6kV005172 for ; Thu, 25 Aug 2005 10:11:07 -0400 To: Sonny Rao In-Reply-To: <20050825022411.GA6057@localhost.localdomain> Sender: linux-fsdevel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org I think this is a libsysfs/iprutils issue due to a sysfs change in recent kernels. Install sysfsutils 1.3.0, then grab the latest iprutils package from http://prdownloads.sourceforge.net/iprdd/iprutils-2.0.15.4-1.src.rpm?download. Rebuild against libsysfs 1.3 with --target=ppc and upgrade the package. Restart the following iprutils init.d services: iprinit, iprupdate, iprdump and re-run your test. Brian Sonny Rao wrote: > Hi Brian, > > I recently noticed a severe regression on certain > filesystems for a simple dd create test. The odd thing was that > filesystems which implemented the "writepages" aop along with > "writepage" aop (which all writable filesystems implement) didn't seem > to suffer nearly as much. So, I was thinking that maybe TCQ or > something isn't quite working correctly. > > I have tried the same file creation test on non-local fibre storage > and don't see the degredation there, and we don't see it on x86 or > x86-64 either. > > I noticed the problem in 2.6.13-rc6 but I tracked it down to a patch > that went into 2.6.12-rc5 which looks fairly benign. It was a merge > from Greg-KH's device driver tree into mainline: > > http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=05d3794aa8bd3b2c9f7920a05003c331cdeb75c5 > > I have confirmed on multiple IPR machines that if I revert this patch > from 2.6.12-rc5, I see huge ( over 2x in some cases) improvements in > the dd file create test on the affected filesystems. > > Do you have any idea why this would happen? > > The only suspicious looking part to me is this one from > drivers/base/core.c > > @@ -89,7 +87,6 @@ static void device_release(struct kobjec > static struct kobj_type ktype_device = { > .release = device_release, > .sysfs_ops = &dev_sysfs_ops, > - .default_attrs = dev_default_attrs, > }; > > But I really have no idea what is going on > > I've attached the script I was using for the tests, the filesystems > which regress are ext3, xfs, and reiserfs. Just edit the variables at > the top for block device and mount-point for whatever machine you want > to run on. > > Thanks, > > Sonny Rao