by Aitor_Ibarra » Fri Oct 01, 2010 7:28 pm
I'm not convinced that iperf and hyper-v play nicely together. I would get different results each test, going from mediocre to abysmal and always consistently lower than I was seeing in more real-world tests (e.g. running tests against iSCSI targets, doing network copies etc). This was with 10Gbit/s, maybe it's more accurate with 1Gbit/s.
With hyper-v role installed, even the 'host' is a vm, although it's a more priviledged one. Maybe it wasn't hyper-v mucking around with the network stack, but the QoS driver. It was as if Windows was saying "oh, you're just running an iperf test, you don't really need the bandwidth". I prefer testing against a Starwind RAM disk, which did give me consistent results.
In theory, you should be able to get about 100MB/sec over 1Gbit/sec ethernet. In both directions at the same time.