Oracle FAQ Your Portal to the Oracle Knowledge Grid

Home -> Community -> Usenet -> c.d.o.server -> Re: RMAN backups to NFS OR FS 'snapshots'

Re: RMAN backups to NFS OR FS 'snapshots'

From: Keith <>
Date: 30 Mar 2007 20:46:00 -0700
Message-ID: <>

On Mar 30, 1:49 pm, NetComrade <> wrote:
> For those who run backups to NFS
> a) which device do you use.. with digiliant, and emc ax100, we haven't
> rearched much beyond 25megs per second (that's 9i, 10g with
> compression is faster)
> b) have you ever had corruption problems? we recently had a corrupt
> lvl0, which I thought was difficult to 'achieve' over a tcp mounted
> nfs
> We have also not used snapshots, but we might want to consider that..
> Could someone describe the 'basics' with that? This is the way I think
> I understand it:
> snapshots are taken at predetermined periods (db has to be in backup
> mode during each one?)
> As snapshots get older, they are 'merged'
> But biggest quesion:
> How does rman work with snapshots :) (e.g. what if we want to run rman
> duplicate, w/o impacting production)
> We run EMC CX500's and CX300's, do those offer snapshot technology?
> How pricey :)
> Any document links appreciated.
> Thanks.
> .......
> We run Oracle 9iR2,10gR1/2 on RH4/RH3 and Solaris 10 (Sparc)
> remove NSPAM to email

I run 10gR2 (RAC,ASM) on RHEL 4 and CX500. We use EMC's SnapView clones (not the "snapshots") to replicate the production database. If you have the Navisphere client, the online help is actually pretty useful for understanding clones and snapshots; but essentially for every production LUN, you map a target, clone LUN. Also check out They have some good white papers on snapping an Oracle database. They're method has several steps, but is flexible.

For my purposes, I just shutdown my target database and start a clone (of my data and flash recovery disk groups). You do not have to do anything with your production database. When the synchronization is complete, I fracture the clones, start asm, and rebuild the parameter file (keep my "old" values with the exception of the new db name, control files, undo, etc).

Then I start the database and let it go through recovery. Then I just use nid to change the dbid and db name, tweak the parameter file one more time, and open resetlogs. This essentially replaces the rman duplicate for me. Is this what you were asking? I'm kinda confused how rman duplicate would impact production. Any overhead incurred on the rman target (of a duplication) has got to be almost nil.

The emc method, if I recall correctly, has you clone the data disk groups first, then the flash disk groups. Then you can to PITR, if desired.

Once your initial setup is done, recloning becomes very easy. For my environment I have it scripted... and it's fast.

I've done it similarly on 9i with IBM's flash copy in the same way... just make sure your target db pfile is consistent with the source, flash your datafiles, controlfiles, redos and start 'er up. Received on Fri Mar 30 2007 - 22:46:00 CDT

Original text of this message