Will spanning tree allow/support point the use of multi-point wireless links for redundancy

Question

I've got a network, with a shared wireless link, I want to add the links identified by the "+"s. After reading up on spanning tree protocol I'm concerned the shared wireless link could be a problem.

           A
           |
         (WAP)--
          /|\   \
        /  |  \    \
      /    |    \     \
   (WAP) (WAP) (WAP)    \
     |     |     |     (WAP)
     B+++++C+++++D+++    |
                     ++++E+
                    (WAP)  +
                      |    +
                      F++++

I have control over the brand of switch, but not the WAPs. *WAP Info Added:* The WAPs are a combination of Tranzeo TR-5a-Nf and UBiQUiTi NanoStation5; specifically the WAPs at A, E & F are TR-5aNf, and the WAPs at B, C & D are NanoStation5.

The redundant links are standard Cat5E copper and connect as shown by the "+" lines in the diagram.

Any helpful suggestions or pointers would be greatly appreciated!

Backstory: The current network is the diagram without the "+" links. B,C,D, etc. (there's actually five points) are on a farm spread out over about a 15,000 sqft area with about 1,000' between point A and the closest point and 2,000' between point A and the farthest point. Individual links go down occasionally (equipment blocks them, damaged, etc.) and we're trying to route around failed wireless links automatically.

I don't know all that much about spanning tree, but when I did some research it seemed to indicate that the redundant links had to be on separate ports and in this scenario they will all be connected to the same port on the root hub (A) by virtue of the multi-point wireless network.

Answer 1

After you have added your links, you will end up with something with the same issue as a mesh network: You have several wireless path toward a destination and you need to use the "best" and forget about the others. "best" being defined by some metric (capacity, loss rate, latency/jitter ...) which

• is not trivial to retrieve
• is invasive, you need traffic to get them
• can change dramatically over time.

This kind of problem require a very specific solution. Your best bet would be to use something like 802.11s or another mesh routing protocol that support mixing wireless and wired links.

Using a wireless-unaware protocol like STP is asking for trouble. If a wireless link degrades but remain usable enough for STP, you will have horrible performance.

Alternatively, if your hardware does not support any mesh technology and you are not willing to change it, i would set A as a 4addr wireless client, so it can try roaming to the 'best' AP. But i admit i haven't understood the situation with E and F.

Answer 2

My answer is platform-independent. Your model can be achieved. You want the wireless and wireline to back each other up (likely the wireless as primary network path with wireline).

Spanning Tree interfaces will carry a cost that the network will use to calculate a loop-free topology; any remaining loops will have their interfaces turned down. Two things to mitigate this are choosing the interfaces to 'normally' be down by adjusting their cost to make them appear less appealing in the topology calculation.

You can also limit the interfaces that give out Spanning Tree packets, or how they announce them (are they an end-link, a trunk/uplink, etc.).

Find the capabilities of your gear network-wide, and likely find another provider that has worked out a similar solution in their environment. Perhaps you will need to break down the L2 flat network into L3 sections joined by an IGP like OSPF or simple static routes, etc.