From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jevon Qiao Subject: Re: Seek advice for using Ceph to provice NAS service Date: Thu, 24 Sep 2015 22:21:58 +0800 Message-ID: <56040706.1000902@gmail.com> References: <56020CC2.2060806@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from mail-pa0-f44.google.com ([209.85.220.44]:34793 "EHLO mail-pa0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756467AbbIXOWI (ORCPT ); Thu, 24 Sep 2015 10:22:08 -0400 Received: by padhy16 with SMTP id hy16so74680082pad.1 for ; Thu, 24 Sep 2015 07:22:08 -0700 (PDT) In-Reply-To: <56020CC2.2060806@gmail.com> Sender: ceph-devel-owner@vger.kernel.org List-ID: To: sage@newdream.net, "ceph-devel@vger.kernel.org" Cc: ceph-users@ceph.com Any comments or suggestions? Thanks, Jevon On 23/9/15 10:21, Jevon Qiao wrote: > Hi Sage and other Ceph experts, > > This is a greeting from Jevon, I'm from China and working in a company > which are using Ceph as the backend storage. At present, I'm > evaluating the following two options of using Ceph cluster to provide > NAS service and I need your advice from the perspective of stability > and feasibility. > > Option 1: Directly use CephFS > Since Ceph as a unified storage can provide file system storage > service via cephfs, this looks an ideal solution for my case if CephFS > is ready to be used in production environment. However, based on the > previous discussions on CephFS, I see that there are still some issues > like not ready for supporting multiple metadata servers, lack of a > fully functioning fsck and so on. Also, I learn that CephFS has been > evaluated by a large community of users and there are production > systems using it with a single MDS from the official website of Ceph. > So it is difficult for me to make the decision on whether I should use > it. > > Option 2: Ceph rbd + NFS server > This might be a common architecture used in current NAS storage. But > the problem is how to get rid of the single point failure on NFS > server. What I have right now is to use Corosync and Pacemaker(the > typical HA solution in Linux) to form a cluster. It seems that > Sebastien Han has verified the feasibility. > > Your comments/advices would be highly appreciated and I'm looking > forward to your reply. > > Thanks, > Jevon