Queue Depth - EVA

[avik]

I wondered is there a way to setup the Q-Depth parameter in EVA for diskgroup or lun?

[Uwe Zessin]

What are you trying to acomplish?

I've been told that the EVA can process up to 2048 outstanding I/Os per controller port, but there is no way to limit it to a disk group (which can hold multiple virtual disks managed by both controllers, by the way) or a single virtual disk.

In the past, the queue full handling on storage arrays coupled with the handling in the Fibre Channel drivers on the hosts weren't too great.

If I rememeber, VCS V3.020 started to offer new handling, but it was no longer compatible with Windows NT. There was a special hack that could turn the array back into old mode, but more details were never published.

[avik]

uwe, i guess the number of disks in a dg has a direct relationship with the
que-depth parameter you setup at the host end.

lets say you hv a dg with 12 disks in it and the host LUN Queue Depth is at 8 that comes out to be 12*8/12 = 8 IOs per drive and lets say a drive can handle only 2 IOs in a queue the rest of the IOs get queued up in the controller leading to slow throughput from the storage. I dont think u can limit the que depth parameter for a lun using the hba driver configuration file ?

In that case either u hv to work backwards and limit the number of luns for the host to address the performance or limit the performance issue back to tht host itself by limiting it frm storage.

pls post wht u think.

[Mark Poeschl_2]

Avik -

See this document on calculating queue depth per LUN presented to a host. Note that the method of 'tweaking' queue depth is very OS-dependent.