Dependency-Driven Task Scheduling and Layer Management for Device-Edge-Cloud Systems

Abstract

The execution of DAG-based applications in Mobile Edge Computing (MEC) systems faces significant challenges due to dynamic resource availability, heterogeneous container image layers, and their tight coupling with task scheduling decisions. Existing DAG- or priority-based schedulers typically rely on fixed dependencies or static priorities, and fail to adapt to runtime variations in computing and storage states at the edge. This paper studies task scheduling in a Device-Edge-Cloud (D–E–C) architecture and formulates the objective as minimizing the makespan of DAG applications. To this end, we propose a resource-aware dynamic priority mechanism that integrates DAG dependencies with real-time computing and storage conditions. Building on this mechanism, we develop a batch-driven scheduling framework that jointly optimizes task offloading, concurrent image-layer loading, collaborative layer reuse, and storage-aware layer deletion under limited MEC resources. Simulation results show that the proposed approach consistently achieves lower makespan and better scalability than conventional DAG- and priority-based methods in dynamic MEC environments.