HP AutoRAID 12H performance

[Maarten van Maanen]

We use an Oracle 8 database on a HP9000 K370-system with a 12H with 6 18Gb HDD's.
Using arraydsp -r a get the message:
Add another disk because the disks are being overworked.

I/Os are stacking up in the command queues for the disks. This creates a
bottleneck which can be solved by adding another disk. The new disk will
take some of the workload off the existing disks and improve performance.

When using arraydsp -r I get similar results:
Vendor ID = HP
Product ID = C5447A
Array serial number = 00000013C8F1
Performance Metrics
--------------------------------------------------------------------------------
Requested Period = 09:00 07/06/2000 to 12:00 07/06/2000
Actual Period = 09:15 07/06/2000 to 12:00 07/06/2000
Interval = 4

Out-of-Spec Indicator = *
--------------------------------------------------------------------------------
Drive Relocate Working Cache SCSI

Difference Blocks Set Util Q

07/06/2000
09:15 1.000 0.000 0.570 0.000 *107.000

10:15 1.000 0.000 0.590 0.000 *111.000


This would mean I/O-request are stacking up in the SCSI-queue. With 6 HDD the SCSI parameter should not be higher than about 30. In practice it's always around 90-110. I am always getting results like this ever since the Oracle has been in production-use.
I'm using both 12H-controllers with 96Mb each and they are cross-linked and both are used.
Is there an other way to improve performance ?

Maarten van Maanen
Netherlands

[Bob Inglis]

You are on the right track. Autoraid performance improves significently up to 8 drives. Above 8 drives only slightly.
Best performance is acheived with 12 drives installed. Also keeping about 10% capacity unassigned improves performance. Beyond that it's time to share the load with another Array.

[Maarten van Maanen]

Hi,

Already figured out that adding disks would help. Last weekend I reconfigured the system using ALL available space (64Gb in this case). System is still building up it's RAID 0/1 space at the moment (currently standing at 7.5Gb) which used to be about 22Gb before. Other disks will be add later.
However, command queue length has not changed a bit. I also get the feeling that tools like SAR, Glance and Top don't give me the total picture. They all seem to be looking primarily at the first disk on the primary path of the AutoRAID since these tools show only give a real workload on that first disk of the AutoRAID.
I still wish the AutoRAID had a bit more fine-tunable parameters...

[CHRIS ANORUO]

Check these links for more info:
http://my1.itrc.hp.com/cm/QuestionAnswer/1,1150,0x15677e990647d4118fee0090279cd0f9,00.html

http://www.hp.com/essd/model12H_writews.html

[Peter Jansen]

Bad performance on a 12H can be caused by a change of parameter:

WRITE CACHE ENABLED

This kills write performance (the OS thinks the writes are unsecure and adds a force flag to each write request).

The only way to get around this is to set the 12H in High Perfomance mode (WRITE CACHE DISABLED can not be set). Command: arraymgr -J HighPerformance

Best Regards

Peter Jansen
ITchallenge