Reliability-aware task-driven serverless edge computing function deployment

doi:10.1016/j.comnet.2025.111673

Return

Reliability-aware task-driven serverless edge computing function deployment

2025-09-06

PRE

DOI：10.1016/j.comnet.2025.111673

Original

Original request for help

Save

Abstract

En 中文

Serverless edge computing (SEC) has emerged as a promising paradigm to enhance the efficiency and scalability of multi-access edge computing (MEC) systems by enabling dynamic and on-demand deployment of lightweight functions. Although the on-demand and distributed execution of SEC systems enhances system performance and resource utilization efficiency, it introduces specific reliability challenges, including unpredictable task assignment, increasing complexity in resource scheduling, and heightened risk of function deployment failure. In addition, dynamic network conditions further exacerbate these issues, posing significant threats to overall service reliability. Accordingly, we propose a reliability-aware, task-driven function deployment framework for SEC systems that jointly optimizes user-to-base station associations, transmission power allocation, and function deployment strategies to satisfy user reliability requirements while minimizing long-term system cost. The formulated problem is a non-convex, nonlinear optimization problem with integer variables and sequential dependencies, which makes it difficult to solve directly using traditional optimization methods. To tackle these difficulties, we model the problem as a Markov decision process and design a proximal policy optimization (PPO)-based deep reinforcement learning approach to find an effective solution. In addition, a greedy algorithm is applied to refine the actions, ensuring better solution quality and performance during the optimization process. Extensive numerical results demonstrate that our PPO-based method satisfies user reliability requirements and reduces cost compared to baselines, highlighting its effectiveness for SEC networks in dynamic environments.