From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S261463AbTJMF2y (ORCPT ); Mon, 13 Oct 2003 01:28:54 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S261464AbTJMF2y (ORCPT ); Mon, 13 Oct 2003 01:28:54 -0400 Received: from thebsh.namesys.com ([212.16.7.65]:24527 "HELO thebsh.namesys.com") by vger.kernel.org with SMTP id S261463AbTJMF2w (ORCPT ); Mon, 13 Oct 2003 01:28:52 -0400 Message-ID: <3F8A3813.7070503@namesys.com> Date: Mon, 13 Oct 2003 09:28:51 +0400 From: Hans Reiser User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4) Gecko/20030624 X-Accept-Language: en-us, en MIME-Version: 1.0 To: "Ihar 'Philips' Filipau" CC: Linux Kernel Mailing List Subject: Re: ReiserFS causing kernel panic? References: <3F89D567.6080901@softhome.net> In-Reply-To: <3F89D567.6080901@softhome.net> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Ihar 'Philips' Filipau wrote: > Hans Reiser wrote: > >> reiserfs is not warranted to work on corrupted hdds..... > > > Is there any kind of error statistics for hard drives? > > Geometry is known. > I suspect that structure of damages, caused by contact of plates > surface with head, can be classified. > > > It may be possible to classify manufacturing glitches. I think HD > producers have this kind of classification/statistics - to improve > quality, keeping price low. > > Actually what I'm thinking of: some kind of design rules for file > systems, how to minimize crashing due to hdd glitches. > Let's say, if some of hdd regions are know to be more error prone - > desing fs to use those regions less. > If hdd damages used to have some specific structure - design file > system to keep renundant data in regions which are less likely to be > lost both at the same time. So renundancy would make sense. > > Is there any thing like this? > > Or file systems now do outlive hard drives?-) > Block allocation policies affect performance a lot, and keeping them simple is important. I would however be interested in knowing what the distribution function for errors by geometry is. If it turned out that, say, errors were higher at the platter edges, I could make some format changes.... I think that for users it is best to think about how to mask drive errors in the device layer or the device using RAID and mirroring. -- Hans