From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Greaves Subject: Re: RAID5 question. Date: Sun, 07 Aug 2005 12:40:47 +0100 Message-ID: <42F5F33F.8020905@dgreaves.com> References: <17137.43778.476908.961140@cse.unsw.edu.au> <01f001c598fd$c59a51a0$0400a8c0@LocalHost> <17138.32835.190046.8088@cse.unsw.edu.au> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <17138.32835.190046.8088@cse.unsw.edu.au> Sender: linux-raid-owner@vger.kernel.org To: Neil Brown Cc: djani22@dynamicweb.hu, linux-raid@vger.kernel.org List-Id: linux-raid.ids Neil Brown wrote: >Technically: no. >raid5 itself doesn't do readahead. > > Could you clarify this line please: >The filesystem or VM layer does read-ahead. > > suggests that either: performance should only be affected by the readahead setting of the block layer immediately below the filesystem or: lvm and md use the VM layer to access the lower block levels and so performance is affected by readahead settings for all involved layers/devices. I strongly suspect the latter as I run xfs over lvm over md over sata and I found (I posted the test script aeons ago) that varying the readhead on different layers affects performance. The best performance (for my usage) was obtained by setting readahead to zero on all layers except the topmost (lvm for me). Comments? David