JET: Decoupled Placement and Online Coevolutionary Scheduling in Data-Intensive Mobile Edge Computing

Abstract

Data-intensive Mobile Edge Computing (MEC) demands a synergy between server placement and task scheduling, yet these processes operate on disparate optimization horizons. This paper proposes JET, a two-stage optimization framework that bridges this gap through placement-aware scheduling. In the offline stage, an evolutionary algorithm optimizes base station placement to minimize deployment costs while maximizing geographic coverage. In the online stage, a distributed coevolutionary algorithm (CEA) performs real-time scheduling and resource allocation across the fixed infrastructure. JET introduces three key innovations: 1) DAG-aware genetic operators that identify critical paths to maximize task parallelism and reduce infeasible solutions; 2) adaptive population sizing that reduces computational evaluations by approximately 30% without quality loss; and 3) dynamic reference point adaptation to maintain diversity across the Pareto front. Extensive evaluation using real-world workloads (Alibaba-Cluster, TPC-H, and Montage) demonstrates that JET achieves up to a 10.42% reduction in latency and 31.61% energy savings over state-of-the-art methods. Crucially, JET maintains zero deadline violations under the evaluated workload and delay conditions while explicitly accounting for realistic communication overheads, ensuring scalable performance in dynamic, large-scale MEC environments.