Design of Distributed Computational Offloading using Ray Framework

Multi-access edge computing has been widely considered as an important technology to support low latency services by reducing application execution latency. In this article we design a distributed computation offloading problem which is solved by using multi-agent deep reinforcement learning. In this scenario, every UE acts as an agent and it tries to maximise its utility in every round of game. The utility of a UE is given as weighted combination of number of processed bits and energy consumed in achieving it. We designed a custom multi-agent scenario for this simulation on Ray platform and trained it using deep reinforcement learning algorithms. The simulation performed showed that the agents reach a stable mean reward.