Multi-agent-Based Load Balancing in Mobile Edge Computing

Since mobile devices have limited resources, they can offload computationally heavy tasks to the edge of the network via a mobile edge device. Recent years have seen a lot of focus on the development of an optimal offloading approach for deciding whether a job should be completed locally or at an edge device. Systems as they stand are inadequate for practical usage because they were primarily developed for scenarios involving a single mobile device or a single edge device. This work presents offloading rules that minimize task drop rates and execution delays in scenarios where there are many non-cooperative mobile devices and multiple edge devices, without requiring knowledge of the environment's dynamics. Mobile devices are only able to partially satisfy this non-cooperative resource allocation dilemma. We show how a reinforcement learning-based strategy can be used to reduce the effects of partial observability and gradually understand the dynamics of the environment and its long-term repercussions. The suggested algorithm dramatically decreases task drop rates while minimizing energy and computing costs, making it preferable to preexisting offloading schemes.