11.31 SCSI queue_depth

[Francis Noël]

Hello forum people

I am digging around to try and make sense on SAN performance and this came up...

I have a number of rx3600 hosts running 11.31.
SCSI devices on hosts connected to an EMC Symmetrix DMX-3 array show "queue_depth = 1" and a host connected to an IBM DS8300 show "queue_depth = 8" when interrogated with the "scsictl -a" command.

Is the queue_depth adaptive under 11.31 ?

HBAs are the same and both the DMX snd DS hosts run B.11.31.0803. Only difference is that the DMX stuff uses HPUX11i-VSE-OE and the DS host is under HPUX11i-OE-Ent. Might that be it ?

[Turgay Cavdar]

You can set/see your max_q_depth with scsimgr command.To see the current value:
scsimgr get_attr -D /dev/rdisk/diskx -a max_q_depth

For example to set to 64 type:
scsimgr set_attr -D /dev/rdisk/diskx -a max_q_depth=64

The default value of max_q_depth is 8. In your situation i think someone mıght just changed the value to 1.

This command also shows the driver default value:
scsimgr get_attr -N /escsi/esdisk -a max_q_depth

[Francis Noël]

Hi turgay

Thanks for the scsimgr wake-up call. I always forget to use that.

All max queue_depth values are still defined at 8. esdisk is default 8 current 8 but the SAN devices are default 8 current 1. I get the feeling the EMC is somehow telling the hosts to use 1. We are looking into it from the SAN side.

[Turgay Cavdar]

Hi Francis,
I dont think that EMC is telling the hosts to use 1. I think someone manually set it to 1.
May be one possibility is fibre channel driver may have some predefined values for different storages, what is your FC card? Also 1 is very low value and can affect the performance badly.

[Francis Noël]

Well it would not be the first time that array surprised us. Further poking has revealed that all 11.31 hosts connected to the EMC via multiple HBAs use a queue depth of 1. All 11.31 hosts connected to the EMC via a single HBA do have a queue depth of 8. EMC's commitment to support 11.31's native multipathing has always been....iffy. We are currently investigating port config on the EMC side. I'll let you know what we find.

[Turgay Cavdar]

We already have 11.31 that connected to DMX-3 with two paths and they all have a queue depth value of 8.

[Francis Noël]

FA ports on the DMX array are hard-configured to allow a queue depth of 1.....in some instances. Other FA ports are set at 8, others are not set at all. Will take up the issue with the manufacturer's support (sigh).