ECO-RL-ECA: Efficient Computation Offloading using Reinforcement Learning in Edge-Cloud Architecture

With the development of numerous delay-sensitive applications, it has become very difficult to perform all the tasks locally because our mobile devices have less computation power. Traditional cloud servers have high computational power and storage, but they’re located far away from IoT devices, so they take extra time and energy for transmission before the execution of tasks. Local devices and cloud servers are not appropriate for these time-sensitive applications. To resolve this real-time issue, the idea of edge computing was developed, where we can offload our task to edge devices that are closer to the user device than a cloud server and hence consume less time and energy to execute the same task. The task can either run locally, on an edge device, or on a cloud server. We modelled our problem from a constrained real space to a finite discrete space using the Markov Decision Process and derived our solutions using reinforcement learning to reduce the execution time for delay-sensitive applications. The proposed method is found to be more effective at reducing execution time.