I've recently isolated the cause of some network performance issues I've had for some time. It seems that network performance decreases drastically when the Hyper-V role is installed on a pair of Server 2008 R2 boxes. This first became apparent to me when I recently installed Intel X520 10G adapters and didn't see performance increase much. After the upgrade I did see occasional peaks of ~1.5 Gbps which made me think everything was working, but I was never able to realize much of a performance increase when doing file transfers.
Since then I have pulled the adapters and started testing them on my bench using different hardware. What I have found is the following:
Intel NetEffect CX4 10 Gbps Adapters (SMB Push): 4.4 Gbps (Avg, No HV) down to 1.05 Gbps (Avg, HV)
Intel X520 10 Gbps Adapters (SMB Push): 5.5 Gbps (Avg, No HV) down to 1.05 Gbps (Avg, HV) (Avg, HV)
Broadcom 8100 Series 1 Gbps Adapters (SMB Push): 0.99 Gbps (Avg, No HV) stays the same at 0.99 Gbps (Avg, HV)
Something I have noticed is that CPU utilization stays about the same while throughput is much reduced with the hypervisor installed. I have to wonder if there is really just this much overhead when the hypervisor is active. Note that this problem occurs even when the NICs aren't part of a virtual switch--it's just the hypervisor itself being installed and active that is enough to cause the problem. No combination of advanced adapter configuration settings seem to make an appreciable difference, but if someone has a magic cocktail of parameters I'd still like to hear them.
The real kicker is that once Hyper-V has been installed, even uninstalling it doesn't return performance to pre-HV role installation levels. They stay stuck at ~1.05 Gbps.
My question I suppose is this: Is this behavior typical? If not, what might be done to improve the situation?