Online Dependency-aware Task offloading in Cloudlet-based Edge Computing Networks

The demand for low-latency processing in mobile devices is dramatically increasing. However, mobile devices inherently lack the capacity to handle heavy and low-latency processing. Edge computing techniques, which offload the tasks of user applications to a nearby server, are being used to mitigate this problem. Moreover, applications requiring rapid processing have evolved, becoming more complex. Application tasks are no longer merely computational and independent; each task requires its associated libraries and dependencies, all of which must be considered during offloading. The dependency between tasks should also be taken into account. Centralized decision-making for offloading for a large number of users is not practically feasible. In response to this issue, we designed a distributed method based on deep reinforcement learning. We defined the states of the learning agent in a manner that enables users to learn about the environment with respect to the level of task dependency. Through simulation, we demonstrate that the proposed algorithm surpasses existing benchmarks in terms of application completion time by identifying the optimal server for offloading dependent tasks.