From mboxrd@z Thu Jan 1 00:00:00 1970 From: David McBride Subject: Re: Ceph Ansible Repo Date: Thu, 06 Mar 2014 13:28:30 +0000 Message-ID: <531877FE.6040106@cam.ac.uk> References: <53186512.6080405@cam.ac.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: Received: from ppsw-40.csi.cam.ac.uk ([131.111.8.140]:41781 "EHLO ppsw-40.csi.cam.ac.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751602AbaCFN2c (ORCPT ); Thu, 6 Mar 2014 08:28:32 -0500 In-Reply-To: Sender: ceph-devel-owner@vger.kernel.org List-ID: To: Gandalf Corvotempesta Cc: Sebastien Han , "ceph-devel@vger.kernel.org" , Weil Sage , Patrick McGarry On 06/03/14 13:19, Gandalf Corvotempesta wrote: > 2014-03-06 13:07 GMT+01:00 David McBride : >> This causes the IO load to be nicely balanced across the two SSDs, >> removing any hot spots, at the cost of enlarging the failure domain of >> the loss of an SSD from half a node to a full node. > > This is not a solution for me. > Why not using LVM with a VG striped across both SSD ? > I've never used LVM without raid, what happens in case of failure > of a phisical disks? The whole VG is lost ? Yes. A stripe-set depends on all of the members of an array, whether managed through MD or LVM. Thus, in a machine with two SSDs, which are striped together, the loss of *either* SSD will cause all of the OSDs hosted by that machine to be lost. (Note: if you want to use LVM rather than GPTs on MD, you will probably need to remove the '|dm-*' clause from the Ceph udev rules that govern OSD assembly before they will work as expected.) Kind regards, David -- David McBride Unix Specialist, University Computing Service