Reinforcement Learning-Based Optimization for Mobile Edge Computing Scheduling Game

Abstract

Task scheduling on edge computing servers is a critical concern affecting user experience. Current scheduling methods attain an overall appealing performance through centralized control. Nevertheless, forcing users to act based on a centralized control is impractical. Hence, this work suggests a game theory-based distributed edge computing server task scheduling model. The proposed method comprehensively considers the mobile device-server link quality and the server’s computing resource allocation and balances link quality and computing resources requirements when selecting edge computing servers. Furthermore, we develop a time series prediction algorithm based on IndRNN and LSTM to accurately predict link quality. Once Nash equilibrium is reached quickly through our proposed acceleration scheme, the proposed model provides various QoS for different priority users. The experimental results highlight that the developed solution provides differentiated services while optimizing computing resource scheduling and ensuring an approximate Nash equilibrium in polynomial time.