-
Notifications
You must be signed in to change notification settings - Fork 132
Possible challenges in mesh formation
An important limitation during mesh formation is that each node can only connect to one AP at a time. As a result a mesh will generally only have node available that can connect to other nodes.
Due to this constraint, there can e a problem forming a complete mesh when it is spread out. For example, if we have nodes A to D, where A can reach B, B can reach C etc., but A can't reach C and D can't reach B, we could get B connecting to A and C connecting to D, but then A couldn't connect to C or D (because they are too far away). Similarly D wouldn't be able to connect to A or B and no complete mesh will be formed. The current solution for this is that B (and C) will keep scanning and will (randomly) decide to disconnect and connect to the other mesh. After, for example, B connects to C, now A will be alone and can connect to B, resulting in a full mesh. As mentioned this is done randomly, because there is no way a node can know if connecting to the other sub mesh will be beneficial for the mesh as a whole or detrimental.
This should work, although due to the random nature, it can be slow, especially if many more nodes are involved. In this case it can happen that one node decides to leave one submesh and connect to the other, but then a bit later another node leaves the second submesh to connect to the first, in essence undoing the previous change. An additional complication can be when A is a AP only node (e.g. a bridge node). In that case A can't reconnect to B, after B connects to C. So instead B needs to switch back to A and then C needs to decide to connect to B, then D would connect to C, finally resulting in a full mesh.
Introduce an anchor (master) node, causing the sub mesh that contains such an anchor node never to try to reconnect. In the above example the bridge would be an anchor node and as a result B would never decide to disconnect from this node. Similarly, when C connects to B it will also be anchored in place and wouldn't try to restructure the network. The existence of the anchor node will be (optionally) included in the node sync message.