Layer-Reused Collaborative Scheduling for Container-Dependent Tasks in Industrial Real-Time Systems

Renowned for their lightweight and portability, containers are increasingly deployed in edge computing to deliver low-latency and privacy-preserving computing services for industrial real-time systems. While layer reuse can potentially improve the execution efficiency of container-dependent tasks, the existing work suffers from two issues: 1) inefficient disk space utilization, which compromises image pull efficiency and reuse utility, and 2) inadequate consideration of resource capacity, leading to practical limitations. This paper addresses the container-dependent task scheduling problem in edge computing-assisted industrial real-time systems by introducing a novel Layer Reuse-based Collaborative Scheduling (LRCS) framework. First, to ensure practical applicability, we comprehensively consider resource capacities and the cost terms of the task completion time. Second, a framework that collaborates scheduling and execution nodes in a weakly coupled manner is proposed. At the execution nodes, LRCS maximizes disk utilization by evicting layers only when disk space is insufficient. The eviction is modeled as a 0-1 knapsack problem, where dynamic layer value assessment enhances the reuse rate. At the scheduling node, LRCS formulates container-dependent task scheduling as a multidimensional online bin-packing problem. A value-based learning algorithm optimizes long-term processing efficiency by adjusting the scheduling order of online tasks. Experimental validation on a testbed using real device data and comparisons with state-of-the-art layer reuse-based algorithms demonstrate that the proposed algorithm outperforms all baselines.