From: Eric Sandeen Subject: Re: [PATCH] e2fsprogs: don't set stripe/stride to 1 block in mkfs Date: Wed, 06 Apr 2011 10:00:40 -0700 Message-ID: <4D9C9C38.2070008@redhat.com> References: <4D9A17F8.4000406@redhat.com> <4D9B45AB.8000208@redhat.com> <4D9B49A6.7000709@redhat.com> <4D9B95DE.6050408@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Andreas Dilger , ext4 development To: Zeev Tarantov Return-path: Received: from mx1.redhat.com ([209.132.183.28]:54929 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752949Ab1DFRAq (ORCPT ); Wed, 6 Apr 2011 13:00:46 -0400 In-Reply-To: Sender: linux-ext4-owner@vger.kernel.org List-ID: On 4/5/11 11:51 PM, Zeev Tarantov wrote: > On Wed, Apr 6, 2011 at 01:21, Eric Sandeen wrote: >> Zeev, if you'd like to test that patch above with your profiling, that'd be awesome. > > My wall-clock timings are strange and I disregard them. > On a desktop with X turned off and perf recording system-wide, I trust > the profile to show everything. Agreed, the wallclock time was just a very coarse sanity check while off at a conference :) > In the no-stride case, I don't think you've made it any worse. it shouldn't even go down this path so I hope not ;) > In the stripe=stride=1, it went from: > > [ perf record: Woken up 13 times to write data ] > [ perf record: Captured and wrote 3.163 MB perf.data (~138211 samples) ] > # Events: 14K cycles > # > # Overhead Command Shared Object > Symbol > # ........ .............. ..................... > ....................................... > # > 13.08% gzip gzip [.] zip > 12.65% flush-253:0 [kernel.kallsyms] [k] ext4_mb_scan_aligned ... > to: > > [ perf record: Woken up 12 times to write data ] > [ perf record: Captured and wrote 3.091 MB perf.data (~135042 samples) ] > # Events: 13K cycles > # > # Overhead Command Shared Object > Symbol > # ........ ........... ..................... > ..................................... > # > 15.10% gzip gzip [.] zip > 14.70% gzip gzip [.] treat_file.part.4.2264 > 11.10% flush-253:0 [csnappy_compress] [k] snappy_compress_fragment > 6.12% md5sum md5sum [.] digest_file.isra.2.2089 ... > 0.23% tar [kernel.kallsyms] [k] ext4_mb_scan_aligned ... > So that's great, but I still don't want mke2fs to set a stripe and > stride of 1, because that's silly - no code improvement in the > allocator is ever going to be able to use that information. I totally agree, and I'm sure the mkfs patch will go in as well. Thanks for the verification of the kernel patch, too. It should help a bit for sane stripe-sizes, too, with less impact as the stripe size gets bigger. Thanks, -Eric >> Thanks, >> -Eric > > Thank you! > -Z.T.