FruityMesh Algorithm

Introduction

A mesh should be able to build and manage all connections without user interaction. There is one big restriction when it comes to BLE connections. Having more than one connection as a peripheral can degrade mesh performance and lead to connection losses. Using one connection as a peripheral and up to three connections as a central has proven a good configuration for mesh connections. With these settings, the following will lead to problems where two nodes are not able to connect to each other because their one connection as a peripheral is already taken.

impossible

There is no way to know the size - or the participating nodes - of a mesh in advance. Distributing the presence of a node over long distances would be a bad idea because of the time it takes and the energy it costs. This means that every node can only see its surroundig nodes.

Simulator

There is a Simulator available that illustrates how the algorithm works. It allows to apply different configurations and see how the mesh reacts.

Node

Each node saves a few variables like network ID, cluster ID, cluster size and its own ID. Additionally, it keeps some information for each of its connections.

saved values

During discovery, it broadcasts its nodeId, clusterSize and clusterId along with a few other measures in special advertising packets (JOIN_ME>> packets). It also scans for some time to receive discovery packets of surrounding nodes. From these packets, it selects the best connection partner and establishes a connection to it.

Clustering

The real trick here is the clustering. Because every node knows the size of the cluster that it’s part of, it uses this information as a criteria when connecting to others. Big clusters can always decide to which nodes they want to connect and smaller clusters will have to obey. Any change in cluster size due to connection or disconnection is broadcast through the exisiting connections.

cluster forming

After a few JOIN_ME packets have been collected, they are processed in the ClusterScore function to determine the best connection partner. A node never tries to connect to a node with the same clusterId to prevent loops.

Handshake

After receiving the JOIN ME packets, some time has passed and the other node might already be in another cluster or in a different state than before. This is why the two nodes will first do a handshake after connecting to pass the latest information to each other. Only one Handshake must happen at a time to prevent race conditions. Once they are satisfied with their partner, the connection is used, otherwise it is disconnected.

States

The algorithm uses a state machine that switches between different DISCOVERY and HANDSHAKE states. This helps the node to reduce its energy usage.

Self-Healing

Once a connection in the cluster breaks up, the smaller cluster distributes a new cluster ID among its nodes. This repairs the missing connection using a similar path.

Self healing

After each connection loss, a node will increment its connection-loss-counter. This counter is used together with the nodeId to generate a clusterId. This is necessary because there might be cases where the original node of a cluster can’t join the cluster anymore if it generates the same clusterId again.

Connection

Master Bit: Because a node can’t always know for certain if it is part of the bigger cluster, there may be times when two connected nodes both think they are part of the bigger cluster. This would pose a threat to the mesh once the connection drops. Therefore, each connection is assigned a masterBit that is passed to the node that is part of the bigger cluster. With this, it is only possible that the masterBit is in transition during a disconnect and both cluster must dissolve, but they can no longer form any islands.

More

Anybody interested in a much more detailed explanation of the algorithm is welcome to take a look at The Algorithm in Detail page.