We recently stood up a new environment, utilizing iSCSI instead of HBAs for LUN presentation. We are, running into (expectedly) performance issues. We've ironed out a few kinks (multipathing via 3 NICs, round robin algorithm, jumbo frames though we've not tested those completely yet, etc). That being said, I noticed that our NIC utilization hovers at around 40-60% when utilizing round robin as opposed to a single NIC policy (fail-over, lack of MPIO, what have you) which sits at around 90% which is ideal. In theory, we'd like to see near the same performance with 4 NICs in a MPIO config to a single 4Gb HBA. Is that feasible? Does the MPIO overhead preclude this amount of throughput? Is there something I'm missing that will pump more IOs?
Thanks.
|