DAG-based Dependent Tasks Offloading in MEC-enabled IoT with Soft Cooperation

Xiaobo Zhou,Shuxin Ge,Pengbo Liu,Tie Qiu

DOI: https://doi.org/10.1109/tmc.2023.3328333

IF: 6.075

2024-01-01

IEEE Transactions on Mobile Computing

Abstract:Multi-access edge computing (MEC)-enabled Internet of Things (IoT) has become a powerful solution to run computation-intensive applications on end devices. These applications are composed of multiple dependent tasks, which can be abstracted as directed acyclic graphs (DAGs). Moreover, applications can share partial intermediate data with each other based on dynamic network conditions to boost its performance, so-called soft cooperation. However, it is quite challenging to make optimal offloading decisions with external dependency between tasks of different DAGs introduced by soft cooperation, as well as the subsequent huge continuous solution space caused. In this paper, we propose a DAG-based dependent tasks offloading method with soft cooperation in MEC-enabled IoT. First, we formulate the problem as a Markov decision process (MDP), aiming to minimize the application latency and energy consumption, and to maximize the cooperation gain simultaneously. Then, we propose a branch soft actor-critic (BSAC) algorithm to make optimal decisions under dynamic network conditions, including the offloaded tasks, the CPU frequency of end devices, and the sharing ratio of intermediate data. Specifically, BSAC uses multiple branch networks to reduce the solution space. Finally, a series of simulations are conducted to establish the superiority of the BASC algorithm over state-of-the-art solutions.