Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S264022AbUDTWNK (ORCPT ); Tue, 20 Apr 2004 18:13:10 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S263836AbUDTWNI (ORCPT ); Tue, 20 Apr 2004 18:13:08 -0400 Received: from eurogra4543-2.clients.easynet.fr ([212.180.52.86]:15037 "HELO server5.heliogroup.fr") by vger.kernel.org with SMTP id S264314AbUDTUAq (ORCPT ); Tue, 20 Apr 2004 16:00:46 -0400 From: Hubert Tonneau To: linux-kernel@vger.kernel.org Subject: The missing RAID level Date: Tue, 20 Apr 2004 19:50:07 GMT Message-ID: <045P8FJ12@server5.heliogroup.fr> X-Mailer: Pliant 91 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2570 Lines: 51 Assuming that I want to do long term archiving (many many gigabytes of datas, but tiny load) on disks, the cheapest and easiest solution nowdays seems to connect large 300 GB IDE disks through USB 2 to a comodity PC. Now the problem is how to best recover from some disks failure ? For the production storage, I use software RAID 5 on internal disks, but for huge external ones, it might no more be a good idea, because of two reasons: . having several disks failure is more likely . in case of a catastrophy, I would like to be abble to recover at least some of the datas, so no RAID at all is better that RAID 5 since I have DVD backups and what I want to optimise is operator time So, one very interesting possibility would be to have an extra RAID level that would do the following: assuming that you connect 5+1 partitions, then you get 5 md partitions, not a single one, with the following properties: . any read to mdX goes straight forward to reading the underlying partition. . any write goes staight forward to writting the underlying partition, but also updates the parity on the extra partition. So, at the expense of slow write capabilities, which is not a problem for long term archiving, I get a system with very interesting properties not covered by existing Linux software RAID levels: . in case of one disk failure, I can plug a new one, then rebuild just as with classical RAID . in case of more disk failures, I only loose part of the archives (so spend less operator time for recovery from DVD). . all partitions can be read just through ignoring the RAID details, so it is possible to unplug any of disks and connect it sowehere else with no extra constrains . adding or removing disks from the raidset is trivial: just rebuild the parity partition On the 'use what's available instead of requesting new features' side, I'm also interested with feedback from users using large (8 to 16 SATA disks) external cheap (anything that raises the price for 8 disk from 8 x 350 euro to more than 16 x 360 euro is no solution since clustering is then the way to go) towers that would make RAID 5 a resonable solution, and how they connect to the Linux kernel (each disk seen individualy, RAID handled by the contoler, need for a driver outside the stock kernel, etc) Regards, Hubert Tonneau - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/