|
Depends on the storage being used and the number of spindles. I
definitely wouldn't go any smaller than 8k, but 8k is probably fine on a
JBOD array, especially if you're matching your filesystem block size to your
oracle block size to your stripe size. You might want to look at changing
it to 16k if you're using high-speed disks and let the disk firmware's
read-ahead and large cache give you a bit of a boost. You're probably not
talking about a huge jump, though - only if you're really saturating your disks
will you notice a difference.
As
I've said in the past, the rules change completely when you're using a "smart"
array like a Hitachi lightning or an EMC symmetrix, or even a smaller one like a
Clariion or HP EVA. For those, engage your vendor's performance engineers
or hire someone like me to come in and tune your storage, or just invest a lot
of time in learning how your array works and experiment.
You
should look at using VxFS as well - you can eliminate double-buffering and get
asynch kernel i/o, which alone could provide much better throughput.
Depending on your Solaris version, as well, you could be creating unnecessary
memory pressure by using UFS.
Thanks,
Matt
--
|
Title: Message
- stripe size AK
- Re: stripe size Matthew Zito
- RE: stripe size Roger Xu
- Re: stripe size Matthew Zito
- Re: stripe size AK
- RE: stripe size Jesse, Rich
- RE: stripe size Odland, Brad
