Re: Oracle database mirroring questions

Sony Antony wrote:
> We have been looking at the database redundancy problem from a high
> level point of view, for the purpose of architecturing our distributed
> application. I m not an expert in Oracle. So I decided to ask them
> here before deciding on any particular solution.
>
> 1. Does Oracle provide any database replication/mirroring mechanisms,
> so that there will be a warm backup database to take over, if the
> main/primary crashes.

Yes - it's actually 'hot', not warm in Oracle speak. It's called DataGuard. Real Application Cluster (or RAC) is a cluster solution where the crashing of one instance does not affect database operations.

> If it does what will be the performance hit (
> cpu usage ) on the primary because this duplication mechanism is
> running in the background.
>

Low. It will be a low profile file read and transmit to another site (Hot Standby).
None for the RAC solution - although it will have overhead as compared to a single instance database.

> 2. If such a mechanism as above exists, how does it work. Does it
> duplicate only the modifications of the original database. Does it
> locks the whole tables while it is duplicating the data. Or does it
> lock just the rows - in which case the client applications will not
> 'feel' its presence much

It reads Oracle internal files (redo log files), that record transactions anyway. As such, there's no locking of tables, and the application(s) are not affected.
http://download-west.oracle.com/docs/cd/B10501_01/server.920/a96524/c01_02intro.htm#54284

>
> 3. Does Oracle provide any clustering mechanisms for fault tolerance,
> wherein N number of machines will all have exactly mirrored databases.
> One can afford to lose a number of its node machines. As long as at
> least one node is up all client applications can run without any
> problems.

Yes - above mentioned RAC
>
> 4. Does Oracle provide any clustering mechanisms for load balancing,
> wherein data will not only duplicated among the nodes as 3. above, but
> different clients can connect to different node machines in a load
> balancing fashion, and they will all see the exact same data.

Yes - RAC in combination with a cunning network configuration

>
> 5. I was of the understanding that Oracle uses raw disk space.

That's an option

> IOW in
> the case of Solaris, it doesn t use the default Solaris UFS file
> system, but uses raw disk space with indexing etc implemented with
> respect to the physical disk location. This makes disk access faster.
> Did I get this wrong. Or is there an option to do it on top of the
> file system or as a raw file system.

I have seen Oracle data files reside on journalled file systems, logical volumes, and raw devices. All working fine. You want all disk performance you can get? Go raw, bypass the OS. Want some level of redundancy? Use the file system of your choice.
>
> 6. One of the possibilities we thought about so as to implement a
> clustering, was to have an NFS server, whose disks are mounted in a
> number of different machines. Each of these machines will run an
> Oracle database server, but accessing the same NFS mounted database.
> Is this possible. I personally didn t think so since if multiple
> machines are modifying the same data at the same time, this will
> result in data corruption, since each machine's modification might not
> be inside a single atomic write() system call.
>

AFAIK, that is possible but not advised. NAS/SAN seem OK (options vary), but NFS has given troubles in the past. Never seen it in use. What if the NFS server isn't up?

Don't know if you are hardware bound, or O/S bound, but several hardware vendors offer complete RAC solutions. Oracle installed, etc. I know (in Europe, at least) that Dell and HP/Compaq deliver such solutions for test purposes. Maybe contact your hardware vendor?

-- 
Regards, Frank van Bortel