Raw Vs JFS

[Rushank]

Hello,

I'm bit confused over this issue.

We are having over 250 GB of oracle Database.
All these databses are spread across many Vxfs file system. I was reading some of the posts related to this says oracle perform better on Raw device file system.
Can any body throw some light on this Please..??

[A. Clay Stephenson]

The only way to know with your data and applications is to measure for yourself. On 10.20 and 11.0, if you use the OnlineJFS mount options delaylog,nodatainlog,convosync=direct,mincache=direct for datafiles and indices and delaylog,nodatainlog for archive and redo logs you will see near optimum performance. The =direct mount options bypass the buffer cache and you are in effect using raw i/o. In fact, I've not been able to measure any significant difference between those mount options and true raw io. I can tell you that if you use vxfs, I've always found the best stripe size for the LVOL's to be 64k because vxfs writes in 64k chunks regardless of block or extent size.

If you are running 11.11, I suspect that using cooked files (no =direct mount options) will be your best performer; again, the only way to know is to measure for yourself.

[Tom Maloy]

The issue is "raw" versus "cooked". In raw, Oracle manages the buffering and disks. In cooked, the operating system manages the buffering and disks.

Raw used to offer a speed advantage, but the OS and hardware vendors have improved their disk performance. Cooked (VxFS in this case) should perform at least as well as raw.

Stick with VxFS. Excellent performance, and much easier to recover after a crash. (Can't fsck a raw disk...)

That's the high-level view.

Tom

[Leif Halvarsson_2]

Hi

Perhaps this can interest you:
http://veritas.com/products/category/ProductDetail.jhtml?productId=oracleeditionhp

[Rushank]

Thank Clay.

Well OS is 11.00 and Oracle 8.1.6. Mount options for these filesystems are just simple with delaylog. (For ex: /dev/vg08/lvol2 /u91 vxfs delaylog 0 7 )
I've never used mincache or convosync etc with any of the file system. Is it OK to use this option to mount these FS??

[A. Clay Stephenson]

You bet. Just do a shutdown of the database, a umount of the filesystems, change /etc/fstab, and then mount the filesystems and startup the database. In 32-bit OS's, this is normally very good because you can reduce the size of your buffer cache and use a larger SGA.

[Rushank]

Can I use this new mount option for all the Oracle File system?? I've few file system that has just oracle orchives and backup directory. The backup directory mainly used during hotbackup.

Is this option correct..?

mount -F vxfs /dev/vg02/lvol6 /u90 vxfs delaylog,mincache=direct convoysync=direct 0 4

[A. Clay Stephenson]

You could but performance is best when archive and redo logs use the buffer cache. You will see the best performance if only the data files and index files bypass the buffer cache. I also suggest that you add the nodatainlog option to the others but especially in the file systems that contain the data files and indices.

[A. Clay Stephenson]

I suppose that I should add if you have archivelogs, redologs, data files, and indices in the same filesystem and you wish to make the data and indices bypass buffer cache but everything else use buffer cache AND you don't want to change anything in Oracle, you could use symbolic links to move the archive directory and redo logs to another filesystem.

[Rushank]

Clay,

Thanks for the reply.

You mean to say I must use mincache,convosync,delaylog and nodatainlog for all the Oracle file system except for the file systems that has archive and redo logs...Right??

And for these FS Can I use just normal mount like..option /dev/vg02/lvol6 /u34 vxfs delaylog 0 4