Real World Oracle Database Backup and Recovery

[GraceO]

What is the real world solution to backing up Oracle databases? We are using HPUX version B.11.23. The Oracle databases are versions 8, 9i, and 10g. Currently, we are using a remote agent for unix servers to get the Oracle DB exports backed up using Backup Exec for Windows servers. If we have a disk failure, we can bring back HPUX because we are using Ignite-UX to backup the OS, but the Oracle 'stuff' will have to be re-built from scratch. Isn't there a better way? Your expertise would be greatly appreciated!

[A. Clay Stephenson · Accepted solution]

Well, the real world method is RMAN. However, since you mention HP-UX, let me mention a Plan B that I have used for years.

1) Shutdown the database.
2) Using OnlineJFS snapof= mounts, create snapshots of your Oracle filesystems. (This takes only seconds regardless of the the size of the filesyetm.)
3) Restart the database.
4) Backup the snapshot mounts.
5) Umount the snapshots.

Total downtime ~ 2 minutes.

You get almost all the uptime of a hot backup combined with the safety of a cold backup.

In Windows, this same technique can be done using the Open File Manager or Volume Shadow Services.

Finally, you should never have to use Ignite to bring you back up from a disk failure because all your OS disks should be mirrored. Disk replacements are then done "on the fly". I have not shutdown a box to replace a disk in something like 10 years although I have replaced many tens of disks during that period. Of course, you should still make your Ignite images because Ignite protects you from your own stupidity (e.g. rm -r *).

[GraceO]

I tend to forget that my drivs are mirrored, thanks for reminding me. But in regards to my own stupidity, I've been working with HPUX since January of this year and let me tell you, I feel a lot better with Ignite in place!

Getting back to Oracle DBs, where can I get more information on the OnlineJFS steps? I am not familiar with OnlineJFS other than it allows you to extend your file systems without having to do a umount. Is there a cheat sheet on the snapof=?

[A. Clay Stephenson]

Actually a "man mount_vxfs" explains it but here's a thread that explains how they work and the syntax very well. It's a bit long and the stuff about co-ordinating backups across multiple hosts can be ignored but pay attention to the mount -F vxfs snapof= stuff plus the explanation of the snapshop mechanism.

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1019274

Typically, you choose a snapshot buffer device something like .15 the size of the original filesystem. That should suffice for all but the most active database applications. "Active" in this context means many inserts, updates, and deletes. Selects don't matter.

[Eric Antunes]

Hi Grace,

I recommend using 2 strategies simultaneously:

- Cold backups (Oracle DB shutdown immediate)

- Hot backups with rman (requires DB in archivelog mode)

Best Regards,

Eric Antunes

[Yogeeraj_1]

hi,

Concerning your Oracle database DATA, I would recommend doing HOT RMAN disk backup. (level 0 then level 1 with Incremental Merge option). The disk backup files you can afterwards move to secondary media.

With Oracle 10g, it is much easier and it has more options that you can use to verify that the backup is OK.

If you need any further clarifications, please let us know.

kind regards
yogeeraj

[GraceO]

Thanks for the advice everyone! I will have to research all options you have presented but at least I will be looking at working solutions.