A Latency-Optimal Task Offloading Scheme Using Genetic Algorithm for DAG Applications in Edge Computing

Edge computing technology presents an opportunity for embedded devices with limited computing power to effectively process even the most complex of applications. Scheduling tasks to the edge server for processing can effectively reduce task execution latency for the user device. However, current task offloading approaches overlook the unique topological relationships and scheduling within tasks by treating user device tasks as a single entity, leading to the underutilization of computing resources. In this paper, the fine-grained task offloading problem is addressed by considering the offloading with precedence constraints among tasks. This approach enhances task parallelism between edge servers and user devices by allowing tasks to be offloaded and executed on different processors. However, this also makes the problem more challenging since the task scheduling sequence and decision-making process become more complex. A lightweight and efficient offloading decision is proposed in this paper for the single-server scenario. This decision enables scheduling of multi-user tasks in sequence, choosing the most appropriate location for execution. The approach was then extended to a multi-server scenario, where the optimal server for each user device is determined using Genetic Algorithm(GA) optimization techniques, resulting in the minimum average computational latency of the task. Experimental results demonstrate that this approach outperforms existing schemes in terms of task execution delay and offloading efficiency.