StarWind Software

I am trying to make sense of the results I am seeing when testing out MPIO with Starwind 6.0.4837.

Target Server:
Windows Server 2012
Device - CDP Image, Thin Provisioned, Write-Through Caching (4096MB Cache Size)
Storage - 6 x 1TB WD Red in RAID 10 on PERC 6e (attached to Dell MD1000)
NIC - Intel i340 Quad GB: 10.250.0.10, 10.250.1.10, 10.250.2.10, 10.250.3.10

Initiator Server:
Windows Server 2012
Microsoft iSCSI Initiator
NIC: 2x Intel Dual GB: 10.250.0.11, 10.250.1.11, 10.250.2.11, 10.250.3.11

Server NICs are connected via 4 VLANs in an HP procurve switch - no flow control, no jumbo packets

Here is my issue. When I connect to the target with a single NIC, I see results I would expect (using ATTO Disk Benchmark):


Next, I add another target IP to discovery, enable multipath, verify I see two paths, Round Robin policy and run the same benchmark and see the improvement I would expect by adding another path:


Repeat with a third NIC:


Again, results get better with each NIC... until I add a 4th. After adding the 4th path my results plummet:


I've verified all the settings I can think of, but no matter what, when I add a 4th NIC to the mix, the results are half of what I'd expect with a single NIC. Hoping someone can help me understand why this happens?

Thanks,



EDIT: I apparently didnt check everything. for some reason, the 4th NIC on a quad port card is connecting at 100Mbps, so that definitely is the culprit. Well, typing it all out helped me find the issue in any case.

What queue depth did you configure within ATTO? Did you try other tests (synthetic or real one, like f.e. copy HUGE file)? Is the issue re-producible with 'em?

Sorry, in my edit at the end of the post I mentioned it was a NIC connecting at 100Mbps. That one was throwing everything off. Ended up being a bad cable in this instance. Thanks again for taking the time.

Jeff

So... Can you publish final numbers with a 4 NIC properly working MPIO please?

jsmithv wrote:Sorry, in my edit at the end of the post I mentioned it was a NIC connecting at 100Mbps. That one was throwing everything off. Ended up being a bad cable in this instance. Thanks again for taking the time.

Jeff