LVM Stripe Size / Oracle Stripe Size / FS Stripe Size

[Daniel Estevam Gomes_1]

Hello there. I need some help in HP-UX.
This is my configuration:

16 x Parity Groups RAID5 7D+1P (HDS USP600)

I want to know what's the better configuration for my LVM.

I'm working with the following variables:

FS block size = 8k
Oracle block size = 8k
Oracle multiblock read count = 16
LVM Stripe Size: 8K * 16 = 128K
Number of LDEVs (LUNs) = 16 (1 LUN from each PG)
PE Size: 8k
Number of stripe per Volume = 16

I would like to receive some recommendations about the configuration above. Is this the better one? Do you have any recommendations to me?

I'll apreciate your comments.

Thank you!

Daniel.

[A. Clay Stephenson]

Your PE size is not 8KiB. PE size is measured in MiB and the smallest possible is 1MiB. Your 8k is almost certainly 8MiB. With one extremely dubious exception, PE size has no impact on performance.

Assumming that this is a vxfs filesystem (and despite what you were told in Oracle class), filesystem blocksize has no impact on performance. Vxfs filesystems are extent-based rather than block based and the writes are always done in large chunks. The filesystem blocksize only determines the smallest amount of space to allocate for a file. The default, 1KiB, is actually fine for Oracle but 8KiB is fine as well. Your LVM stripe size is ok. I would not go smaller than 64KiB or larger than 256KiB but you really should do some testing to determine your optimum although 128KiB will almost certainly be "close enough".

I suspect the biggest impact would be to choose another RAID level (RAID 1/0 is almost always the fastest although the least efficient in terms of storage utilization.) The RAID level impact is also dependent upon how cache-centric your array is.