I/O question

[Charles McCary]

Group,

Here's the details of my setup and of what I'm seeing:

1) occasionally gpm shows a couple of luns at 100% utilization.

2) during this time gpm also shows top io processes in a cache state.

3) box is an rp7420.

4) disk array is eva 3000

5) we are using securepath.

6) On HP's advice we did nothing special when building these luns (just created disk group and logical disk and presented them to the host). The securepath path was then utilized to build a vg and filesystem, which oracle now uses. HP stated that eva3000 was "smart" enough to handle hot spots.

Question: Is there a preferred way to layout the db on luns created on eva3000, given securepath and HP's recommendations?

Anything I can check besides what I'm already looking at to give me some further clues?

I appreciate your help.

c

[Steven E. Protter · Accepted solution]

Shalom Charles,

You can take a look at how your application(s) uses the data and look for ways to improve performance.

A common issue is heavy write activitiy on RAID 5 disk arrangements. This is much slower than Raid 10 or Raid 1 configuration. To identify heavy write action you will need to measure performance with glance or sar.

http://www.hpux.ws/system.perf.sh

Another issue is stupid application code. I've recently seen situations where an application found data corruption and kept retrying updates to the corrupted data sequentially until all i/o was dedicated to this task. This requires a close look at how the application works and possibly contacting an internal expert or support.

SEP

[Duncan Edmonstone]

Charles,

Only thing I'd say is, make sure that if you have a couple of LUNs, then each LUN is preferred to a seperate controller. You'll be able to check this with 'spmgr display'.

If the EVA isn't servicing many other hosts you could alos look at increasing the SCSI queue depth for these LUNs (man scsictl for details, search on scsictl and eva on forums for more info)

HTH

Duncan

[A. Clay Stephenson]

I doubt that you actually have a problem. Glance (and any other host-based performance tool) doesn't have a clue that what it sees as a disk device is actually an array LUN which might be comprised of many physical disks not to mention a chunk of array cache. All Glance can know is that a whole heaping helping of i/o is going to that there "disk".

In general, if you are given a choice, choose RAID 1/0 over RAID 5 but with highly cache-centric arrays that doesn't make as much difference as it once did.

[Charles McCary]

Guys...thanks for the replies, I'll look into the things suggested above.

Have a great weekend!

[Charles McCary]

asdfasdf