Queue length EMC disk

[Dimitry Snezhkov]

Hello, everybody.

I am trying to figure out if glance cannot deal properly with
Storage arrays. Here's the thing:

many EMC disks Queue lengths
are shown as over 65000. This
is impossible because performance tools on EMC side show quite the opposite.

I have given up on sar disk data collection long time ago.
But I still had faith in Measureware. Please, explain, what MWA can or cannot do properly in regards to disk arrays like EMC.

I appreciate any information
which will get me closer to understanding of current situation.

Thanks.
Dimitry

[John Bolene]

I would imagine it is the measuring tools on UX.

I know that you can't get a good disk utilization on a VA or XP, the calculation is made for slow disks and does not work for fast ones.

I have an EMC Symettrix (spelling?) on a UNISYS mainframe and consistently get queue lengths of zero or 1, sometimes 2, with access times of between 4 ms for short I/O to 15ms for 60K I/O and this is on a 4 MB channel.

[G. Vrijhoeven]

Hi Dimitry.

Im not sure but i guess the UNIX uses SCSI emulation over the fibre channel. The SCSI queue depth is defauld 8. But the EMC does does not queue per disk ( virual disk for the EMC) , so this explains the difference.

Hope this helps.

Gideon

[George A Bodnar · Accepted solution]

This is a bug we had opened with HP and EMC. EMC suggested upgrading PowerPath which did not fix the problem, but HP offered that a kernel patch had been released for HP-UX 11i (PHKL_27200) and is in the works for HP-UX 11.00.