An Adaptive Event-Driven Task Migration Strategy for Collaborative Mobile Edge Computing

Abstract

Collaboration among edge nodes via task migration is a promising solution for mobile edge computing (MEC) to efficiently meet time-varying and spatially unbalanced computing demands with limited resources. However, existing task migration strategies are primarily confined to a discrete-time paradigm, which suffers from modeling inaccuracy and control dilatoriness. To address this issue, this paper proposes an online stochastic learning-based adaptive task migration strategy operating in a continuous-time event-driven framework. Aiming to ensure satisfactory MEC performance for differentiated services with minimum energy consumption, the policy optimization of task migration jointly with resource allocation and transmission scheduling is formulated as event-driven continuous-time Markov control processes, accurately capturing the stochastic dynamics of task arrival, migration and processing. By leveraging gradient estimation and stochastic approximation, an online stochastic learning algorithm is developed to address the constrained policy optimization. It is computationally efficient and environment-adaptive, capable of real-time computing and online optimization in various unknown environments. Numerical simulations are conducted to evaluate the performance of the proposed strategy, and comparative results confirm its effectiveness.