I'm experiencing some packet loss to all of our VMs that we didn't have before we made some changes to our Hyper-V implementation (Windows 2008 R2). Most of the VMs also run 2008 R2 - with 3 that run Server 2003.
The host server is a Dell R610 with three 4 port NICS - two Intel quad port gigabit and a quad port Broadcom. We us the individual ports of the Broadcom for host management and live migration - no problems here. We use the Intel cards for both iSCSI and VM networks. Calling the two intel cards “A” and “B”, and the ports P1-4 we've used AP1, AP2, BP1, BP2 (ports 1 & 2 of both Intel NICs) for iSCSI connections, and we've created a NIC Team with AP3, AP4, BP3, and BP4 (ports 3 and 4 of both Intel NICs). The team type is "Virtual Machine Load Balancing". We then created a Hyper-V switch based on this team for use with all of the VMs created on the host. (as a side note: prior to implementing the NIC team, we just had 4 Hyper-V switches, one associated with each of these 4 ports.)
The 4 ports of the NIC team are connected to two different Cisco SG200 switches - AP3 and BP3 are connected to switch1, and AP4 and BP4 are connected to switch2 (in an attempt to maximize redundancy). The two Cisco SG200s are simply connected to the rest of our network - each to a different switch within the subnet. There is minimal configuration done to the SG200s (for example NO link aggregation); spanning tree is enable however.
My question is: can the network cables be connected to different switches (as they currently are) and if so is there some configuration piece (either on the switch or within Windows) that I'm missing? What are the options here if this configuration is incorrect? The packet loss is in the range of 0.1%, but we've had odd spikes where a VM was essentially unavailable for a brief period (a few minutes) then returned to "normal" (0,1% loss). Pinging a device (like the SG200 itself) or another physical server (for example our domain controller or the hyper-v host itself) results in essentially 0 loss; maybe one or two packets during the course of a 12 hour ping (this was the “normal” ping response to VMs before we created the NIC team, so I’m quite sure this has something to do with it).
Thanks in advance!