How does MS Unicast NLB work via HUB and SWITCH?

Question

Computer networks

Good afternoon.
I'm setting up a Microsoft Network Load Balancing cluster in unicast mode. This mode works in such a way that the poppy address of the network card on each node is replaced with the same one on all nodes of the cluster and looks like 02-bf-xx-xx... therefore, so that there are no identical poppies on different ports, cluster nodes do not use the poppy address of the form 02-bf when responding, and in the source of the packet they substitute the poppy of the form 02-sequential_node_number-xx-xx... accordingly, the switch cannot add (learn) the correspondence of the port and poppy address of the cluster to the table and sends packets to all ports, which causes performance problems. To solve this problem, it is advised to connect all cluster nodes through the HUB, and already connect the hub to the switch port and thus the switch will be able to add (learn) a poppy of the 02-bf type to the switching table and there will be no unicast storm. The question is what if nodes never respond from poppy address 02-bf, how will the hub help? How does it help the switch learn mac? What is the fundamental difference whether NLB members are connected to a switch and a hub?

Answer 1

And why do you need such old technology?
The servers respond to arp requests with their own mac-address, and the ip and Mac of the cluster are indicated in the packet headers.
And in order not to flood into all ports, the servers must sit in their own separate vlan - the flood will only be within it. Well, or a hub, yes (go look for it again) - then for the switch everything will be on one port.
Well, Microsoft recommends not using unicast in production because of these side effects.