From mboxrd@z Thu Jan 1 00:00:00 1970 From: Anthony Liguori Subject: Re: Hot backup vs (local) live migration Date: Wed, 25 Jul 2007 16:02:59 -0500 Message-ID: <46A7BA83.7050207@codemonkey.ws> References: Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Claris Castillo Cc: xen-devel@lists.xensource.com List-Id: xen-devel@lists.xenproject.org Claris Castillo wrote: > Hi, this is the third time I post this question to the list but I have not > received any response. > > It is not clear to me why the live migration technique used by Xen and > documented at > http://www.cl.cam.ac.uk/research/srg/netos/papers/2005-migration-nsdi-pre.pdf > can't be used to perform a hot backup of a virtual machine. Based on the > paper, just before > shutting down and destroying the original VM, there is an exact clone of it > in the remote/destination host. > Could this technique be leveraged to perform a hot backup? If so, why Xen > does not provide with hot > backup yet? I have been experimenting with VMWare and the performance is > pretty bad since > it uses REDO files. It depends on what your definition of "hot backup is". Live migration while it's active does not result in a consistent machine on the other node. What I mean by this, is that during the live migration process, you could not interrupt the migration and just "run" the VM on the target node (since only a portion of the memory may have been transfered). Now, you could basically keep a tight loop of live migrations that would result in a backup on another node that would exist at a sightlier earlier time than the source node. That is, your backup may be the actual VM 1.5 seconds ago. The danger here is that if you have an application committing transactions to a database, you may end up committing the transaction twice after the backup activates (or something equally bad). So the other option is to have what is know as "lock-step" execution where the hot backup and the source virtual machine are kept in sync at all times. You usually pay a performance cost for this but you do get some pretty interesting fail over. I'm pretty sure there have been more than one research projects devoted to implementing this in Xen. Regards, Anthony Liguori > By the way I tried doing local live migration and I got *several * error > messages. I am not sure if the errors > have to do with the fact that it is not possible to perform a local live > migration or some other technical issues > I can solve if I look at it. > > > I would appreciate any suggestions/advice in this regard. > cc