Creating Oracle dbase on row devices

[Inesa Clinko]

Hi!
Could anybody explain the algoritm /step by step/ to create Oracle dbase,based on raw
devices. And what could be the main principles for creating this base.

Inesa

[T G Manikandan · Accepted solution]

check the attache

[Indira Aramandla]

Hi,

As Manikandan has provided the information regaring what is a raw device and advatages in choosing them......etc, refer to the attachement for a step by step procedure to create and oracle database using raw devices.

Before attempting to create an OPS database with the DBCA, you should have already installed and configured the cluster software from your OS vendor prior to installing the Oracle software. If the cluster was setup correctly, then the Oracle Parallel Server software should have been installed and copied to all nodes in the cluster. You can check this by running the command "lsnodes" which should list all nodes in the cluster.

On many UNIX systems, when a raw device is created, a character device and a block device is created for the same raw partition. Ensure that the character raw device is the name being used.

The device name that must be used with Oracle when specifying a datafile is "/dev/rfoo".

Other platforms, such as Sun Solaris, may have the character and block devices in different directories, for example:

/dev/rdsk/c0t1d0s0 - symbolic link to the character raw device
/dev/dsk/c0t1d0s0 - symbolic link to the block raw device

In this case, /dev/rdsk/c0t1d0s0 must be used when specifying a datafile.

Note: make sure you choose 1M > than the datafile size for the raw device size. Once you defined the raw device you cannot resize, you will need to define another raw device inorder to increase your datafile size.

[Hein van den Heuvel]

> The device name that must be used with Oracle when specifying a datafile is "/dev/rfoo".
:
> In this case, /dev/rdsk/c0t1d0s0 must be used when specifying a datafile.

I beg to differ. I would recommend to never ever give Oracle a direct path, raw device or file!. Just point it to a softlink and
have that point to the raw device.
Much more flexible!

Also... please consider why you want RAW devices and for what parts of the database !
If it is for RAC enabling, well then you may have little choice unless you are on Tru64 with its cluster file system.

If you are enabling RAW for performance you may be able to jut do it on selected database parts, but not all. This allows you to do standard backup procedures most of the way. For example you can create raw very easily just for REDO, TEMP, and UNDO (RBS). Then if you need a backup, you temporarelry re-create those (small!) in the file systems,
do a backup, and re-create raw (big again).

I like RAW for REDO because it can avoid the READ - MODIFY - REWRITE cycle that a file system will do for non-block-sized IO. Redo on raw will do just the writes, no need to pre-read blocks.

Cheers,
Hein.