I have an odd network glitch happening at home. My network consists of an HP Procurve 1810g-24 main switch with a couple of D-Link APs as well. I have a Dell C2100 (FS12-TY as it's also known) for my main server that is running Server 2012 R2 Datacenter and has several Hyper-V VMs running on it. Everything operates as a VM and the host is just a host.
What happens is regularly one of my VMs (my file server, also running 2012 R2 Datacenter) will briefly lose network connectivity. It's most often noticed when watching a video from the file server and it will spontaneously stop playback and I'll have to wait a few seconds before I can restart it (sometimes I can restart it immediately but its been taking a few seconds lately).
The frequency this happens can vary; very rarely can I get though an entire movie without it happening, sometimes it happens once or twice and other times it happens so frequently I get fed up and go do something else. It does this on hardwired desktops on the same switch, laptops attached via one of the APs (desktops/laptops are either Windows 7/8.1), Xbox 360 hardwired, and Android phones (via FTP). It's also done it via hardwired Linux desktops.
The only common element is the network or the VM/VM host. It's hard to tell if other VMs on the same host are having a problem because I don't even notice this unless I leave windows explorer open on a computer (and will see the mapped drive disconnect then reconnect) or am watching a video.
I've looked in the event logs of both the host and the VM and the only thing of significance is that occasionally (and with no discernible pattern) it will lose the RPC connection to one or both DCs (also VMs). This shows up on the host as well (it's also joined to the domain which AFAIK is a supported config for 2012 R2 Hyper-V) but I have not looked at the DCs yet (going to do that next).
I've tried turning SR-IOV on and off on the VMs as well as the virtual switch to no avail (I'm not 100% positive the physical NIC supports SR-IOV, I think it does but SR-IOV also indicated it should fall back to "typical" methods if sufficient resources were unavailable leading me to think it should be safe to leave on). VMQ is currently still turned on and I've heard it can have issues with Broadcom NICs but all of mine on the server are Intel (two onboard and a PRO100/VT quad port addon). The quad port is currently in an LACP trunk (with switch support) but I've also attempted putting the main virtual switch on a single NIC to no resolution.
↧