This seems to be a problem that has existed in all versions of Hyper-V. I have experienced this in the both Windows Server 2008 and Windows Server 2008 R2 versions.
I have two Hyper-V 2008 R2 servers with a couple of Windows Server 2008 R2 guests. If I set up a ping as follows:
ping -t myserver.thedomain
I get solid ping responses, but randomly I will get either a timeout and/or a “network route could not be made” message. This is particularly the case when the Guest OS tries to access another resource via the LAN. The connection does get reestablished, but this causes havoc for any services that rely on a good constant connection i.e. pretty much everything.
The issue seems to involve network interface cards (NICs) and their drivers that do not process Checksum Offloads correctly.
On my servers I have the following configuration:
NIC 1 – Marvell Yukon 88E8056 PCI-E Gigabit Ethernet Controller
NIC 2 – Marvell Yukon 88E8056 PCI-E Gigabit Ethernet Controller
NIC 1 is my management LAN NIC and is dedicated to the virtual server parent. It does not participate in any traffic with the Guest OS’. NIC 2 is my Virtual LAN NIC and is dedicated to the Guest OS’ i.e. the “Allow management operating system to share this network adapter” is un-ticked. I believe this to be adequate best-practice (more NICs would be better for redundancy and performance) as it makes sure you can always access your Host OS even if the Guest VLAN NIC is getting slammed.
NIC 2 (my VLAN NIC) is where the workaround needs to occur. I disabled the following “advanced” networking options:
- IPV4 Checksum Offload
- Large Send Offload (IPv4)
- TCP Checksum Offload (IPV4)
- UDP Checksum Offload (IPV4)
From this list, it appears the issue is to do with the way the Offload is occurring – possibly just for IPV4 but perhaps IPV6 as well. Regardless, as soon as I disabled these settings it resolved my connectivity problems straight away. I also tried the latest driver version of the NIC but that didn’t resolve the issue.
