Changing the queue depth value on L class systems

[Rainer Wolafka]

I am trying to change the queue depth value of disk devices on external L class servers (HP-UX11) using the command "scsictl -m queue_depth=21 /dev/rdsk/cxdytz". When checking the value with "scsictl -a /dev/dsk/cxtydz" I see that the value did not changed at all.

When doing exactly the same commands on a D-class server (HP-UX11) there are no problems and I am able to modify the queue depth value (same storage attached).

(1) Does anyone has a clue why this in not working on L class server and how to change the queue depth value on this hardware ?
(2) Will I see the same behaviour on N and V class servers ?

Many thanks for your feedback.

[Andy Monks]

I believe it's related to the different disc drivers in use on different machines (which basically puts it down to h/w differences). If you check ioscan between your L and D class you'll see they use different drivers.

[Victor BERRIDGE]

I wonder if you havent made a syntax error: queue depth has to be as 2 words:
/usr/sbin/scsictl -m queue depth=21 /dev/rdsk...

Else, I know it works when you have already a existing queue, default=8,
otherwise you need to create and register a start-up script...

Regards
Victor

[Victor BERRIDGE]

Sorry Rainer,
you syntax was correct, I should have checked before, so now I will try to see how I managed the queues but it was some time ago, and keep you informed
Victor

[Alan Riggs]

It may be that the disk in question will not support a queuedepth of 21. In such cases scsictl may return an error but it may also simply make no change to the value.

As a test, try setting the queuedepth to a smaller value and see whether the setting is retained.

[Victor BERRIDGE]

Hi Rainer,
Coulnd get on an L or N class to see, I agree with Alan, your device may not support such value, Have you tried what Alan suggested?
Yet You havent mentionned what queuedepth you have set now, as an alternative, if your queue depth is set to 8 which HP recommends as a place to start, I would give a try at changing the I/O Time-Out parameter, the default value of HP=20 (seconds).
try to set the Time-Out parameter to 60:
1) Verify the I/O Time-Out parameter value using pvdisplay:
pvdisplay /dev/dsk/c0t0d0
PV Name /dev/dsk/c0t0d0
VG Name /dev/r5vg01
PV Status available
Allocatable yes
VGDA 2
Cur LV 9
PE Size (Mbytes) 4
Total PE 994
Free PE 92
Allocated PE 902
Stale PE 0
IO Timeout (Seconds) default # <- Default =>20

2) change the value to 60 using pvchange -t:
# pvchange -t 60 /dev/dsk/c0t0d0
Physical volume "/dev/dsk/c0t0d0" has been successfully changed.
Volume Group configuration for /dev/r5vg01 has been saved in /etc/lvmconf/r5vg01.conf
#

3)verify:
# pvdisplay /dev/dsk/c0t0d0
--- Physical volumes ---
PV Name /dev/dsk/c0t0d0
VG Name /dev/r5vg01
PV Status available
Allocatable yes
VGDA 2
Cur LV 9
PE Size (Mbytes) 4
Total PE 994
Free PE 92
Allocated PE 902
Stale PE 0
IO Timeout (Seconds) 60

#
4) Repeat the steps for each scsi device...

Hope this will be of some help
Victor

[Rainer Wolafka]

Hello all,
I am using the same storage device on the D-Server (on which I can change the queue depth value) and the L-Server (on which I have the problem to change the value). So the Hardware supports this queue depth value (I cheked it).
On both servers I am using the "sdidk" driver (ioscan output). The storage is attached via Fibre Channel (same topology). On the D-Server I am using the "old" HP Fibre Channel adapter while I am using the new one on the L-Servers (A5158A).
"Beside" the systems architecture and the used HBA I see no difference. It should also be possible on the L-Servers to change the queue depth value using scsictl.
Any ideas or tips are welcome. Thanks for any feedback.