Packet Loss during UCS vNIC FailBack

[tka]

Hello

I'm facing the following situation with a UCS infrastructure : when shutting down FI - A  uplink , all vNICs  fail over to the second FI B. At most one ping is lost. When re-enabling the uplink , the failback  occurs but I loose almost more than 10 ping ( about  10s).
I changed the value Failback timer in the windows Ethernet adapter, associated it to a connection policy. which I added to all the vnic templates. after acknowledgment of the action , all the servers rebooted. but I still get the same behavior !
Any suggestions to resolve this ?

[Philip D'Ath]

I'm not familiar with this product.

However if you are not using switched assisted bonding (such as LACP or PAgP), and you failover from one NIC to another, and assuming you retain the same MAC address, the switch is not going to start forwarding packets to the "new" port until it has received a packet from the server so it knows where the MAC address is located.

Worse still if old school spanning tree is enabled on the switch port it will block and go into a learning mode for 30s (usually).  So I hope you are using either rapid spanning tree, or marked the ports as spanning tree edge ports so the switch doesn't place the ports into learning mode at all.

If you do not retain the same MAC address then the server has to send a gratuitous ARP and wait for receiving nodes to receive and accept it before two way communication will work again properly.

[tka]

Hello, thanks for your reply.

Actually there is no Spanning tree. And effectively , the vNIC sends a G ARP as soon as it fails over or back. Wich seems not happening during Failback.

[Walter Dey]

I think you should also consider the OS, and the NIC teaming / multipathing ! UCS is just the hardware of the solution/problem; it will just signal network down to the OS.