Re: UNIX root backup/SAN disk image vs. Oracle Hot Backups

From: Howard J. Rogers <hjr_at_dizwell.com>
Date: Thu, 02 Sep 2004 08:21:42 +1000
Message-ID: <41364b94$0$5288$afc38c87@news.optusnet.com.au>

Frank Foss wrote:

>
> "Ibrahim DOGAN" <idogan_tech_at_yahoo.com> wrote in message
> news:6bf58828.0409010559.5ccd70bb_at_posting.google.com...

>> Can UNIX root backup or SAN disk image backup be reliable as Oracle
>> database backup?
>>
>> I don't know internals about root backup or disk images but it sounds
>> like you may end up with inconsistent control file/datafiles when you
>> load them back .
>>
>> is there anybody out there who knows how root backup/disk image works
>> and can tell whether they can replace Oracle hot backups?

>
> The concept is developed to reduce the downtime for a cold backup.
> A SAN disk image (EMC calls it Business Contingency Volume --BCV)
> is an identical synchronized copy of your live filesystem at a point in
> time. The connection between the two is broken off, and the live
> filesystem goes along doing its thing, while the BCV is static.
>
> Consider the steps of a normal cold backup:
> 1. Shut down Oracle normally
> 2. Start backing up all relevant files.
> 2b. (Wait until backup finishes...)
> 3. Start Oracle
>
> Step 2 can take considerable time. May not be good for application
> availability. Enter the BCV/Snapshot/Image functionality.
> 1. Shut down Oracle normally
> 2. Sync the live fs with the clone. This step should be reasonably quick.
> 3. Break the connection
> 4. Start Oracle
> 5. Continue to back up the cloned fs in peace and quiet. Additionally, if
> the live FS gets trashed, you can snapshot back to the state it had at
> step 2.
>
> I think you are worried about the internal consistency of the files on the
> cloned image. That is correct, the snapshot must be of a cold database.
> This is no different than backing up the original files from their
> original location. You only get the ability to do a cold backup in "no
> time" seen from the Database or application.
>

>>
>> Thanks a bunch
>> i.d.

>
> All the usual disclaimers apply...
> Foz

I've nothing much to add except that if you're going to break a mirror in order to grab a backup, you don't actually have to shut the database down but can instead use the suspend (and resume) statements.

Suspending a database causes it to be checkpointed and then stops all I/O, so you have a consistent, cold backup. But it doesn't destroy the instance, and hence you don't have to suffer through cache warmup performance issues.

You then simply resume the database after splitting the mirror. The backup is effectively a 3-second job, if that, for a terabyte-sized DB.

Now whether that specifically can be done with your EMC BCV...

Regards
HJR
(Another three-letter acronym). Received on Wed Sep 01 2004 - 17:21:42 CDT