Decentralized task coordination in heterogeneous IoT clusters

Abstract

The continuous expansion of IoT devices in various industries and the increasing complexity of such intelligent systems demand a shift from the cloud-centric model to a decentralized IoT architecture. In the new model, IoT devices collaborate locally with each other to relieve remote data centers, latency, and bandwidth. This paradigm shift raises novel research questions, such as decentralized workload and task coordination. IoT devices are autonomously forming heterogeneous clusters in which tasks can be assigned to each other. This thesis considers how this allocation can be decentralized at runtime. The coordination of tasks must consider the available resources and capacities in the cluster. Moreover, it must not assume a static infrastructure because of the dynamic nature of IoT devices. To this end, we have designed and developed three task coordination strategies. As a starting point, we use an approach that decides on randomness which device gets the work, and employ this strategy as a comparison baseline for the remaining two algorithms. The second algorithm is based on Ant Colony Optimization (ACO) and was adapted to our problem so that pheromones are placed along the links that tell how attractive a path is. The third and final strategy was built using a simple gossip protocol. We have conducted various experiments with 10, 25, 50, and 100 devices (VMs) performed in a static and dynamic environment with device outages. We show that ACO finds a matching node with the smallest number of hops and messages sent, while the Gossips strategy can allocate the most tasks successfully. Gossips thus shows the best system reliability, but ACO scales better. The work highlights that ACO performs better than the baseline approach Random. ACO demonstrates a promising candidate for decentralized task coordination in IoT clusters. Our work provides the foundation for further research and advancements in this area.