I am building a Windows Server 2012 R2 Hyper-V Fail Over Cluster based on the following components:
- 3 x Dell PowerEdge R610 (big cpu / lots of ram / 4 onboard nics / additional quad nic)
- 2 x Cisco 3750 ethernet switches (not stacked)
- 1 x Dell PowerVault MD3220i
MD3220i connected with 2 data ports per controller to the first 3750 and two ports per controller to the second 3750. All there ports on the 3750 are put in the same vlan. FlowControl for receive is on. MTU is set to 9000.
The R610 Hyper-V hosts are connected in the same pattern as the storage array. 2 ports on the onboard nic are iSCSI-ports: 1 to the first 3750 and 1 to the second. And 2 ports of the additional quad port mic are iSCSI-ports: and here also 1 to the first, and 1 connected to the second 3750 switch.
All server iSCSI nice and all storage array data ports are configured with the same logic as factory default on the MD3220i. So 4 subnets (in my config 10.5.1.210, 10.5.2.210, 10.5.3.210, 10.5.4.210 for the data ports on the first controller, and .211 for the data ports of the second controller.
The servers are configured in the same logical order (srv1: 10.5.1.10, 2.10, 3.10, 4.10, srv2: 10.5.1.20, 2.20, etc).
The strange thing is when I test a ping with a length below 1480 all ports (server to server, server to array) are working. But when I test ping with a jumbo length (for example 8000) only the first and last data ports on the arrays reply. This is independent of server or controller.
Tests:
- Srv1 10.5.1.10 to MD3220i controller 0, data port 0 10.5.1.210: OK
- Srv1 10.5.2.10 to MD3220i controller 0, data port 1 10.5.2.210: fails
- Srv1 10.5.3.10 to MD3220i controller 0, data port 2 10.5.3.210: fails
- Srv1 10.5.4.10 to MD3220i controller 0, data port 3 10.5.4.210: OK
- Srv1 10.5.1.10 to MD3220i controller 1, data port 0 10.5.1.211: OK
- Srv1 10.5.2.10 to MD3220i controller 1, data port 1 10.5.2.211: fails
- Srv1 10.5.3.10 to MD3220i controller 1, data port 2 10.5.3.211: fails
- Srv1 10.5.4.10 to MD3220i controller 1, data port 3 10.5.4.211: OK
The test pattern above has exactly the same result on srv2 and srv3.
I can ping server to server on all ports using jumbo frames. All switch ports are configured equal. There is an OK and a failing ping on both switches (data ports 0 and 2 are connected to the first switch, data ports 1 and 3 are connected to the second switch)
Any idea on this strange situation?