From mboxrd@z Thu Jan 1 00:00:00 1970 From: keld@keldix.com Subject: Re: md RAID10 questions Date: Tue, 11 Oct 2011 00:15:47 +0200 Message-ID: <20111010221547.GA30555@www5.open-std.org> References: <4E93657F.1090500@meetinghouse.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <4E93657F.1090500@meetinghouse.net> Sender: linux-raid-owner@vger.kernel.org To: Miles Fidelman Cc: "linux-raid@vger.kernel.org" List-Id: linux-raid.ids On Mon, Oct 10, 2011 at 05:37:03PM -0400, Miles Fidelman wrote: > Hi Folks, > > As far as I can tell, md RAID10 is the greatest thing since sliced bread > - mirroring and striping in one operation (more efficient disk use, > better performance), avoids the RAID5/6 write hole issue, etc. - I've > been having very good luck with it on two servers that I run. > > The thing is, I'm looking at some scaling and clustering options, and > can find practically no documentation or case studies or much of any > information about md RAID10 - which sort of surprises me. I'd think it > would be more popular and there'd be more comparisons vis-a-vis other > options. > > Anyway... what I'm looking at is trying to create a large storage pool, > out of 16 drives, spread across 4 servers - essentially trying to create > a poor man's storage cloud to support a collection of VMs that can be > migrated for load leveling and failover. > > One thought is to expose all the drives via AoE, then combine them into > one large RAID10 array - but I'm neither sure this is feasible nor how > it would perform under various loads, failure conditions, etc. > > Thoughts, pointers, ....? there was some bencmarks on our wiki, but it is gone because kernel.org is down. Any ideas how to remedy the wiki problem? Best regards keld