From mboxrd@z Thu Jan 1 00:00:00 1970 From: Zdenek Kaspar Subject: Re: mdadm raid5 with lvm: advantages? Date: Tue, 11 Jan 2011 20:11:45 +0100 Message-ID: <4D2CAB71.8080108@gmail.com> References: <4D2CA32E.3030303@gmail.com> <20110111234945.7c601067@natsu> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20110111234945.7c601067@natsu> Sender: linux-raid-owner@vger.kernel.org To: Roman Mamedov Cc: Richard Grundy , linux-raid@vger.kernel.org List-Id: linux-raid.ids Dne 11.1.2011 19:49, Roman Mamedov napsal(a): > On Tue, 11 Jan 2011 19:36:30 +0100 > Zdenek Kaspar wrote: > >> It makes sense to use LVM for virtualization and iSCSI to get rid of big >> file images (unwanted fs overhead/fragmentation). But yes, in some use >> cases this is OK. > > Since you mentioned fragmentation - there are ways to both make sparse file > images, and to defragment them when needed (on some filesystems like XFS and > btrfs). But when using LVM instead of file images, the user has neither: LVM > can't have sparse LVs, and it can't (easily) defragment an LV that is > fragmented over one or several PVs. Or am I missing something here? Yes, some filesystems can solve fragmentation issue. But if there's no good reason to have images directly on fs, then no need to waste performance (even if it's just 5%). I think you have options to move PE's, but it's tricky and it's better to avoid such allocations in first place.. HTH, Z.