Multi-Subtask Offloading for Edge Intelligent Controllers in Industrial Internet of Things

Edge Intelligent Controller (EIC) is widely used in Industrial Internet of Things (IIoT) as the core component of Industrial Edge Computing (IEC). Intelligent IIoT typically involves massive computationally intensive tasks, multi-subtask of resource-constrained EICs are offloaded to the edge server (ES) for execution, thereby obtaining lower computing latency. Therefore, this paper investigates the multi-subtask offloading (MSTO) problem in IEC system, where there exist diverse offloading scenarios with varying parameters, such as the number of tasks, transmission rate, and edge computation ability. Accordingly, multiple tasks with dependencies are modelled as directed acyclic graphs (DAGs). Then, we develop a sequence to sequence (Seq2Seq) model and Asynchronous Advantage Actor-Critic (A3C) collaborative algorithm for MSTO (S-A3C) to obtain the best offloading policy. The performance analysis and comparisons demonstrate that our method has better decision-making, outperforming several other baseline algorithms. In conclusion, this paper provides a new perspective on reducing the overall computation latency of multi-subtask IEC system in IIoT.