Hi Iavn,
I guess that in your diagram SW1 and SW2 are running STP but SW3 and SW4 are not.
First, especially if you use third party devices, I want to warn you that there is no standard defining what "disabling STP" means, so interpretation may vary. In Cisco's PVST, we flood BPDUs, which allow you indeed to break the loop between SW1 & SW2 (SW1 and SW2 just see redundant point to point links between them).
Your configuration will however -1- be less efficient that running STP everywhere, plus -2- it breaks one of the basic asumption of STP.
-1- When a link fails between two hosts that don't run STP, the reconvergence can only be based on timer. In your example, if the link between SW3 and SW4 breaks, you will need max_age + 2xforward_delay to recover with STP. Even with RSTP your convergence time will depend on timer, which is not efficient.
-2- But the real problem, which is related to the same scenario is that when the network has converged after the failure of the link SW3-SW4, bringing back up this link will result in a temporary bridging loop. This is because SW3 and SW4 are not running STP and will put their ports directly to forwarding. Even temporary, loops are bad in a L2 network:-(
At last, backbonefast is absolutely helpless if you don't run STP on each switch. Backbonefast is supposed to save you waiting max_age when a bridge that does not have any alternate port loses its root port. In the case where you only have two bridges running STP (SW1 and SW2), you will never get into this situation anyway. If you run STP on all 4 switches, then it makes sense. As already mentioned by Georg, Backbonefast is a proprietary feature.
I highly recommend running RSTP (whether Rapid-PVST or MST) on all switches if possible.
Regards,
Francois
