FC60 and I/O performance

[UNIXGRUPPEN]

I've a FC60 (256mb cache) and 4 SC10 to store about 110GB of quite I/O intense database data.

The system is a N4000 running HPUX 11.0 and Progress 9.1C.

Now after migrating more and more user to this system i've noticed the disk I/O escalating at high speed.

The 110gb LUN is setup in one logical volume on a total of 14 disks (raid 1/0).

The LUN itself is assigned to controller A in the FC60.

The first performance bottleneck that i saw was that the system had a huge amount of logical disk I/O.
Since the system has got 8gb of RAM and the diskbuffer setting was at 50% that was quite obvious.
The hitrate in the FC60 cache was then at aprox. 1%.

So the diskbuffer was tweaked down to only 5% and the cache hitratio in the FC60 went up to 3%.
The disk I/O is down a bit, but still quite high.

The next step for me is to split the logical volume in to 2 LUN's instead of only 1, and put them on both controller A and B in the FC60.

Is there anything else that I can do to further optimize the performance ?

To expand the FC60 cache to a maximum of 512mb is quite a cost, so that will have to wait until nothing else works.

I've also thought about expanding Progress database buffer somewhat.

Any hints ? :)

Regards,
Johan

[David Bell_1]

Johan,

My suggestion would be to create at least two LUNS per volume group, one on controller A and one on controller B, then alternate the primary path to use both HBAs on the host as well. Well, that is assuming you have more than one HBA. In other words;

paths:

For first Volume Group

c0t0d0 ---> primary
c1t0d0 ---> alternate

For second Volume Group

c1t0d0 ---> primary
c0t0d0 ---> alternate

Hope this helps,

Dave

[Eugeny Brychkov]

I agree with Dave, try 'balancing' LUNs across FC60 controllers. Another thing you can think about is I/O concurrency. Creating one LUN should create one I/O queue within FC60. If you'll create 4 LUNs and merge them into one VG then OS will be able to read them in parallel, thus creating different queues for different LUNs.
That's just my idea. It can work, but I'm not sure
Eugeny

[UNIXGRUPPEN]

Thanks for all the answers!

Making it 2 LUN's and split them one each on both controllers is already scheduled :)

The I/O is about 75% read and 25% write.
Will it help to adjust the FC60 cache to more read-cache ?

Is there anything else to tweak that have slipped my mind ?

Regards,
Johan