Re: RMAN or Hot Backup

From: jgar the jorrible <>
Date: Thu, 26 Mar 2009 11:07:14 -0700 (PDT)
Message-ID: <>

On Mar 26, 9:59 am, John Schaeffer <> wrote:
> We run RMAN.  We have some people here that are fighting to stop using
> RMAN and move to Hot Backups.  I prefer RMAN, but I'm curious on other
> opinions.......

The only reason to use hot backups is if you have an expensive hardware solution such as triple mirroring, that can split off a mirror more rapidly than other ways of moving data around.

The reason you have to put Oracle data files in backup mode for a hot backup is that Oracle blocks contain several OS blocks. If you use an OS utility to copy the files, it becomes possible to copy a file while Oracle is writing an OS block, and before Oracle writes the next OS block. This is called a fractured block. In order to avoid this, hot backup mode writes additional data to redo logs because additional recovery will be required when this is detected in a restore. RMAN is smart enough to detect this during a backup, so you don't need to write this additiona redo - it simply retries a few times until the the write of the subsequent OS blocks are done.

In addition, RMAN has already built-in lots and lots of housekeeping of all these backups, and the ability to do incremental backups, and the ability to do block-level restores. Yes there is a learning curve, but it is far less work than reinventing all those wheels. Going from RMAN to hot backups is going backwards, bigtime.

Of course, if you are heavily investing in hardware, it may be possible to simply use the RMAN on a downstream replicant site, if that is what you have in mind.

If you are preparing for disaster scenarios with a distant database, Dataguard is much more sensible, because it doesn't have to ship so much data, just a redo stream, and it's smart enough to handle a fair number of network issues automatically. You can RMAN that downstream, as well as locally, so any recovery can be done either place as necessary.

I know that hardware people are heavily pushing these newfangled cheap snapshot SAN's, I'm not at all convinced it is a good idea compared to what Oracle has wrung the bugs out over years. But I understand how IS managers can be pissed about arbitary Oracle licensing costs versus hardware they can hold and touch and virtualize and reconfigure disks from database to document server on the fly. What I don't know is whether I can tell these newfangled thingies that I'm using 8K blocks, so keep them together and consistent from Oracle's point of view. I would be surprised if I could. Anybody know? Anybody know a good way to force a replicable block fracture during a hardware snapshot demo?


-- is bogus.
Received on Thu Mar 26 2009 - 13:07:14 CDT

Original text of this message