1836973 Members
2673 Online
110111 Solutions
New Discussion

SCSI Queue

 
SOLVED
Go to solution
blal
Frequent Advisor

SCSI Queue

Hello All ,

What exactly a SCSI queue ,with reference to LVM and LUNS created from EMC Symmetrix/Calriion.

Does every LUN has its own SCSI Q,so there by adding more LUNS (and thus more Qs) ,am I getting any performance advantage .

Is SCSI Q is a property of VG or a LUN or SCSI Card ,or for the Array itself ?

What is a default SCSI Q for a VG with One LUN,what command will list it ?

Can some one please give more details .
Live and let live.
1 REPLY 1
Solution

Re: SCSI Queue

blal,

>>What exactly a SCSI queue ,with reference to LVM and LUNS created from EMC Symmetrix/Calriion.

For an overview of SCSI queues (or to give them there proper title SCSI tagged command queues) can be found here:

http://en.wikipedia.org/wiki/Tagged_Command_Queuing


>>Does every LUN has its own SCSI Q,so there by adding more LUNS (and thus more Qs) ,am I getting any performance advantage.

Yes, every LUN has its own Q. You *might* see a performance improvement having more LUNs, it very much depends on your IO profile. Certainly having more LUNs allows you to have more outstanding IOs. Of course you could just increase the Q sizes on your existing LUNs and acheive largely the same result (assuming the LUNs are carved from the same set of disks on the disk array).

>>Is SCSI Q is a property of VG or a LUN or SCSI Card ,or for the Array itself ?

It's a property of a LUN. Howvere you need to understand the maximum number of outstanding IOs per disk array port as well in order to make sure you don't exceed it.

>>What is a default SCSI Q for a VG with One LUN,what command will list it ?

The default Q is 8 on HP-UX, there is a kernel parameter that sets this system wide (scsi_queue_depth IIRC).

You can list and change the Q depth per LUN on HP-UX using the scsictl command:

scsictl -m queue_depth /dev/rdsk/c1t2d0

As stated previously you should only change this up once you have established the maximum number of outsatnding IOs the port on the disk array can handle. You then need to make sure that the sum of the Q depths of all the LUNs presented out of that port don't exceed the maximum number of outstanding IOs, otherwise you could get timeouts.

HTH

Duncan

I am an HPE Employee
Accept or Kudo