HP va 7100 lun and oracle

[Soupault François]

what is the better configuration for use vith an oracle data base : one big lun, or multiple small luns ???
Thanks

[Michael Tully · Accepted solution]

I would recommend at least for HPUX systems that any databases would benefit greatly from having smaller LUNS. How big is the disk array that you may wish to split up? I have a few different disk arrays, but it would depend on the combination of servers or platforms that your going to run on them. If I was running databases only 10Gb LUNS are a good flexible size to choose. They are large enough to use in volume groups for DB's and not too large to waste space.

[Vincent Fleming]

Here's a little background on why you would want more LUNs for a database...

In a word - queue depth.

The queue depth is the number of outstanding I/O's that can be queued to a disk or LUN. All disks or LUNs have a maximum.

So, if your max queue depth to a LUN is say, 32, then having 10 LUNs would give you a max of 320 I/Os queued to the database drives... this adds performance.

On the VA7100, the queue depth is 256, and it is *shared* by *all* the LUNs. So, if you have one LUN, the max queue depth is 256. If you have 10 LUNs the max queue depth is (approx) 25 per LUN, or 256 total... but it's just slightly more complicated that that... the 256 is dynamically shared between the LUNs, so that you can have 10 LUNs, but if you're doing I/O to only one of them, you can still use the 256 on that one LUN.

In a word, the queue depth on the VA7100 is for the whole box, not by individual LUNs.

So, to directly answer your question - it makes no difference how many LUNs you use. Use however many is convienient for you (for example, one for logs, one for indices, and one for dataspaces.).

Good luck!

[Janne Paul]

Don??t know if you have dual fiberchannel paths to the VA?
If you have it??s very important that you will config lvm so the primary link will go thru controller 1 on the VA. If pri link is set thru controller 2 you will have a performance loss inside the VA. This could be fairly big.
This is because all luns belongs to redundancy group 1=controller 1

How to check this:
do armdsp -i, then armdsp -a
check loop id on controller 1. Now do ioscan -fn, check devicefilename for your luns seen by controller 1.
If your lun1 on controller 1 is /dev/dsk/c1t0d1
and the same lun1 thru controller 2 is /dev/dsk/c2t0d1, then put c1t0d1 as your primary link.

It could be wise to ask HP for help on this.

FYI This handled differently on the VA7400.

You might also want to modify kernal buffer cache setting depending how your application works.