From owner-freebsd-questions@FreeBSD.ORG Fri Mar 5 03:01:32 2004 Return-Path: Delivered-To: freebsd-questions@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id DE7D716A4CE for ; Fri, 5 Mar 2004 03:01:32 -0800 (PST) Received: from dire.bris.ac.uk (dire.bris.ac.uk [137.222.10.60]) by mx1.FreeBSD.org (Postfix) with ESMTP id 0AD3A43D2F for ; Fri, 5 Mar 2004 03:01:32 -0800 (PST) (envelope-from Jan.Grant@bristol.ac.uk) Received: from mail.ilrt.bris.ac.uk by dire.bris.ac.uk with SMTP-PRIV with ESMTP; Fri, 5 Mar 2004 11:00:12 +0000 Received: from cmjg (helo=localhost) by mail.ilrt.bris.ac.uk with local-esmtp (Exim 3.16 #1) id 1AzD1V-00016P-00; Fri, 05 Mar 2004 10:57:45 +0000 Date: Fri, 5 Mar 2004 10:57:45 +0000 (GMT) From: Jan Grant X-X-Sender: cmjg@mail.ilrt.bris.ac.uk To: Chuck Swiger In-Reply-To: <40474997.60403@mac.com> Message-ID: References: <20040303222612.W39053@guldivar.globalwire.se> <20040303213641.GA37555@happy-idiot-talk.infracaninophile.co.uk> <200403040050.53556.danny@ricin.com> <40474997.60403@mac.com> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: Jan Grant cc: freebsd-questions@freebsd.org Subject: Re: 1 processor vs. 2 X-BeenThere: freebsd-questions@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list List-Id: User questions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 05 Mar 2004 11:01:33 -0000 On Thu, 4 Mar 2004, Chuck Swiger wrote: > The parity calculations for RAID-5 are a lot of work and that work scales > linearly with the number of drives in the array. The longer you make the > array, the worse the performance becomes for small writes in particular. How did you come to this conclusion? For a RAID 5 with a single parity drive, the reason you zero the disks out completely on initialisation is to set up the integrity of the parity check. Then any update to any RAID5 with single parity requires a read of two drives (the target sector and the corresponding parity drive), an in-memory exclusive or against the new data, and two writes. Reads and writes can be in parallel. The "work" for parity updates only scales linearly with number of disks if you use a naive parity algorithm. Or, obviously, if a drive fails. -- jan grant, ILRT, University of Bristol. http://www.ilrt.bris.ac.uk/ Tel +44(0)117 9287088 Fax +44 (0)117 9287112 http://ioctl.org/jan/ There's no convincing English-language argument that this sentence is true.