oracle metaluns best practice

castleknock,

I did some research on this and I think there are benefits to going either way. One particularly for going with the host based stripe is the ability to place luns across both SPs while with a Metalun they would all be on the same SP. Then at least you could do some load balancing for the application on that device. A couple of downfalls are matching stripe sizes for good performance, you could cause more issues if the wrong luns are striped together, and that the CLARiiON has no clue what luns are connected together on the host. Of course there are many more goods and maybe some more bad, but these come to mind. I bring up the  "CLARiiON has no clue" issue because I have reviewed Navisphere Analyzer files for customers before and sometimes you see devices that have the same IO, but have nothing to do with each other. Later you find that they are connected at the host level, and then more digging is needed to make sure no stripes use luns from the raid group (causes performance issues).

As for going with Metaluns, one major thing from a support standpoint of knowing the configuration.

I found a whitepaper listed under the CX3 arrays which may have some more information: http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/H796-implementing-oracle-emc-clariion-stor-sys-wp.pdfImplementing Oracle on EMC CLARiiON

Under "Other considerations for performance" it talks about host based striping and Metaluns.

The Best Practice Guide also has a section on Volume Managers.

Really it comes down to personal preference I think. Hopefully people who have implemented the setup give their input.

-Ryan

ok I had a chat session and after some initial confusion as EMC  Confidential was written on the documentation and I am convinced the  agent though I was a competitor trying to be smart as I asked could I not get a clarification from whoever wrote the doc. I was told that a a  MetaLUN should not be used and rather a basic LUN. I suspect if I  drilled down into the why's that triple striping is the reasoning, but  the reasoning was left out and I was given a straight this is what we  recommend full stop

I think whoever gave you that answer didn't understand what you asked, as I don't believe that answer is correct. It also sounds like they were a bit rude or just ignorant if they either wouldn't take the time to explain the answer or couldn't explain the answer.

I believe that the recommendation to not use MetaLUNs with Oracle ASM is in the usage scenario where you will present multiple Striped MetaLUNs to a Oracle host and then use ASM to stripe those MetaLUNs on the host. This is the triple striping. However if you are not going to use ASM to stripe the MetaLUNs using the host LVM then the use of MetaLUNs is fine as the EMC document stated.

So you can either manage the striping and expansion on the array or the host, but should not do both due to the stripe upon stripe upon stripe result. It seems to me that a lot of Oracle implementations tend to assign multiple smaller LUNs to the Oracle host and let the Oracle DBAs use the LVM to stripe at the host level. But it can obviously be done either way.

Also check out the White Paper on metaLUNs in the Document section on this forum

White Paper EMC CLARiiON MetaLUNs - A Detailed Review.pdf

glen

Qustion to Gurus. If you present LUNs to a host and use LVM or Oracle ASM to create stripe storage then if there is a requirement to do snapshot or clone of your data is it possible? How do you accomplish that goal?

storage array is oblivious to what you are doing with LVM on the host side so just use your standard cloning technique.

If you do need to have the cloning of luns be an atomic action if the luns contain striped data that the Clariion in unaware of, then you can specify the luns when cloning as an atomic action, so to prevent some of the luns being sucessfully cloned and not others leading to an unknown state when the cloned luns are then presented to ASM or similar.