High collision or drop rate QoS queue question

[AmanaLiving]

Hello

We have a procurve 8212zl with redundant controllers and fabric modules.

We are getting a large number of "High collision or drop rate" messages on our HP lefthand SAN ports. We have a 4 node cluster of P4300 G2 SANs, and a 2 node cluster of P4300 SANs. We currently get higher throughput from the 2 node cluster on older hardware. I think this is because of the high number of drops we are seeing on the 4 node cluster.

I have double checked the settings on both clusters, all ports are set to Auto, with flow control on, on both the Switch and the SAN's.

I have been reading the manuals, and it indicates that QoS is in use by default and by default has 8 queues, and all untagged packets get put into queue 3. When I run a "show interface queues C21" all of the drops are occuring in queue 3.

Each SAN node has 2 NIC's, we currently have the 4 node cluster connected with one NIC of each node to ports C19-C22, and the other nic goes to D19-22.

The 2 node cluster goes to ports A7-A8, and B7-B8. The port configurations are identical to the C and D module ports.

My theory is that the inter node traffic plus the server ISCSI on 4 node cluster is overflowing the queues on each module, resulting in lots of drops, but the inter node traffic and ISCSI on the 2 node cluster doesn't reach that level, resulting in a higher throughput.

I was planning on reducing the number of queues to 4, as per the recommendation in the advanced traffic management guide but I am not sure after reading about QoS Passthrough Mode.

I am leaning towards using the QoS Passthrough mode because theorectically should be able to set the passthrough mode, write mem, and perform a reboot of the standby module, once up, switch redundancy, and reboot the second module without downtime, hopefully.

Any comments would be helpful.