Configuring Oracle on a RAID 5 machine

[anuragmin]

Hi,
I was wondering that since RAID5 creates one logical volume of disk across multiple drives there is no need to have multiple copies of control files and also multiplexing redo logs is not requires,

I might be wrong ... I am not sure.

what care needs to be taken in creating and placing datafiles, redo log files and control files on a RAID 5 machine.

Thanks
Anurag

[rcaballe]

What care? nothing (It's the same), but you are not running a good environment.
Controlfiles in RAID 5? mmmmhh..ok
Online Redo logs? NO...Why?
Because RAID 5 doesn't perfom very well in writes and redo logs are Serial Writes, you don't even enable write cache in online redo log disks because their goal is to recover the database (of course they are then archived).
The people who knows :D say put 2 disk not mirrored to place your multiplexed online redo logs, small disks in different controllers.

[kmesser]

certainly there is far less chance that you'll lose files due to disk failure with a RAID 5 system. However, the performance issue is important. One thing you can do is partition your RAID subsystem into multiple arrays, using one for redo and another for data . . .

You might want to look up OFA on technet.

[anuragmin]

But the partitioning will be logical according to the RAID 5 architecture. The sole purpose of multiplexing control and redo log files is to have them on different physical drives and not logical.

[mbaliga]

Hi,

If you are using storage array (like EMC, IBM Shark) which comes with lot of write cache then it does not really matter. You can place your redologs, controlfiles and datafiles all in the same array.

If you are having convetional raid then it is advisibale to put online redo logs on idependent disks and use ORACLE mirror. Put the archived redo logs on mirrored disks. You can put the data component on raid5.

Baliga

[anuragmin]

Even in case of conventional RAID how does it matter whether you have your redo logs on datafiles on the same array or on different arrays as the volumes are spread across different disks.

How does this help in improving performance . The only gain which I see when you use RAID is the recovery of data.

How does one take care on I/O Contention amongst datafiles in case if RAID is implemented.

[marist89]

Sounds like one RAID 5 volume is all you have. If that is the case, you can't. Although your reads and writes will be distributed across multiple disks, all reads and writes will be on the same device.

[anuragmin]

How does having 2 RAID volumes help. They still will be distributed across the same set of drives .

Or is it that the volumes have their independent set of drives across which they are created.

Pardon my knowledge as I am not much familiar with RAID.

Thanks
Anurag

[marist89]

Yes, when I say multiple RAID volumes, I mean distinct sets of devices, not two logical volumes on the same set of disks.

For example, say you have six disks in your system (sd0, sd1, sd2, sd3, sd4, sd5). If you have one RAID 5 device composed of all the disks, your I/O's will get distributed across all six disks. However, all six disks will spin when an I/O is requested.

If you split the disks into 2 RAID 5 groups, you will have one RAID 5 filesystem on three disks (sd0, sd1, sd2) mounted as /u01 and another RAID 5 filesystem on the other three disks (sd3, sd4, sd5) mounted on /u02.

You would then distribute your datafiles on both /u01 and /u02 filesystems for best performance. Although your I/Os would only be striped across three disks, you could manage your disk contention better.

[hptse]

No matter what storage you are using, if only one control file or one online redo file per group are using, you will have a chance that the rdbms damaged the file.

For those files which has a lot of write access , by multiplexing you may prevent it from corruption by the rdbms.