From mboxrd@z Thu Jan 1 00:00:00 1970 From: Theodore Tso Subject: Re: [PATCH] [RFC] mkjournal: zero journal blocks only when necessary Date: Wed, 29 Jul 2009 13:16:49 -0400 Message-ID: <20090729171649.GB14105@mit.edu> References: <71a0d6ff0905120549h628146d8p3def31b09b79199a@mail.gmail.com> <1248883096-2294-1-git-send-email-alexander.shishckin@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-ext4@vger.kernel.org To: Alexander Shishkin Return-path: Received: from thunk.org ([69.25.196.29]:46121 "EHLO thunker.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750780AbZG2RRE (ORCPT ); Wed, 29 Jul 2009 13:17:04 -0400 Content-Disposition: inline In-Reply-To: <1248883096-2294-1-git-send-email-alexander.shishckin@gmail.com> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Wed, Jul 29, 2009 at 06:58:16PM +0300, Alexander Shishkin wrote: > Will something like this do? > > The only blocks that might theoretically (although very unlikely) be > dangerous for newly-created filesystem's integrity are those that > still contain valid signatures, others can be safely skipped. > > Since reads are generally faster (or at least, not slower), this > gives some performance increase during mkfs run. Did you bother benchmarking what this would do on normal disk drives? Previously we were writing out the blocks to be zeroed in large chunks at a time for speed reasons. This patch reduces it to reading the journal one block at a time, and if it contains a valid signature it writes a zero block. The patch also doesn't check for commit blocks, which are just as much a problem (if not more so) than revoke blocks. - Ted