Re: Exadata backups

From: Riyaj Shamsudeen <riyaj.shamsudeen_at_gmail.com>
Date: Tue, 29 Oct 2013 12:25:20 -0700
Message-ID: <CAA2DszwEk=LdyJBcG4=k0OT_U1DKEFzKmaiQ-3Kf+JH2+vVGWA_at_mail.gmail.com>



I think, performance improvement Alan noticed, is probably due to the IB network. ZFS appliances are directly connected to the IB network (compute and storage nodes are connected to the IB network, as you know). For level 0 backups, rman channels must read the blocks from the cellserv to compute nodes and then write to the ZFS appliance (which is usually mounted as NFS file system). Both read and write calls flows through IB network, however, data flow is: request from compute node(s) -> blocks from cellserv -> compute nodes -> written to ZFS file system, data flow contained in the IB network.

For incremental level 1+ backups, there are some optimization, blocks which do not qualify for backup are not read in to the compute nodes at all, i.e. cellserv does block filtering. Also, (If I recall correctly) the block level change tracking mechanism works at block level in exadata(rather than group of blocks as in non-exadata) and so, blocks without recent activity can be completely eliminated in the cellserv layer itself.

Of course, HCC helps to reduce the size of the database.

To answer your question directly, another advantage with ZFS is that it supports HCC compression. So, you could potentially take a ZFS backup and mount that as a database (but require little bit tweaking to the backups though). But, you can't do that with EMC server.

Cheers

Riyaj Shamsudeen
Principal DBA,
Blog: http://orainternals.wordpress.com
OakTable member http://www.oaktable.com and Oracle ACE Director

--
http://www.freelists.org/webpage/oracle-l
Received on Tue Oct 29 2013 - 20:25:20 CET

Original text of this message