Losing UDP traffic in a Hyper-V environment with NVGRE and can't explain why?

I was working on a project where I had to setup Hybrid connectivity. NVGRE isolated tenant network connected up to an Azure vNET using IPSEC VPN over a HNV gateway. Awesome stuff :-)

After the IPSEC connection was established I deployed a VM in Azure and tested the connection. All looked fine until I tried to domain join the Azure VM to the tenant AD.

I started basic troubleshooting using DNS lookups. Resolve-DnsName resulted in error but Resolve-DnsName -TcpOnly came through perfectly. Other unexplained connectivity issues between tenant VMs which were spread over different Hyper-V hosts where experienced by other project members as well.

As I didn’t setup the fabric, I asked which physical network cards were being used. I was informed that a mix of Broadcom 10G cards where installed. I remembered that Broadcom cards have the encapsulating task offload setting enabled by default and this had caused me issues in the past with another costumer. After disabling this setting on the Hyper-V hosts used with the HNV gateway and the one hosting the tenant DC, DNS lookups where functioning properly. I joined the VM in Azure and verified solid connectivity.

The customer then disabled this setting on all Broadcom NICs on every Hyper-V hosts. The lesson learned here is to test, test, test and do not rely on the default settings of a NIC vendor.

HTH, Ben

comments powered by Disqus