I have two HP DL 380 Gen8v2 servers -- identical hardware. Using Windows Server 2012 R2 DC. Up to date Windows updates and HP service packs. Both have the HyperV role enabled and are joined to a VMM 2012R2 server. I'm using a single logical switch with two uplink ports in a team. Host1 has no issues at all and everything works fine. On Host2, all guests have very sporadic TCP performance -- yet the host is perfectly fine. Sustained transfers (using iperf) go from 3Gbit down to 0 and back up again constantly. RDP is unusable. UDP and ICMP packets have no problem.
I've tried disabling VMQ on the NICs on Host2 even though they seem to not be causing a problem on Host1.
The NIC cards are HP FlexFabric 10Gb 2-port 554FLR-SFP+, configured as 2 x 10Gbit NICs. These are rebranded Emulex cards -- not sure of the Emulex model.
There are no errors or warnings in the Windows App/System event logs.
Does anyone have a clue as to where I should look? Has anyone seen this very odd behavior before?