AI-Driven Multi-Agent Scheduling and Service Quality Optimization in Microservice Systems

Abstract

This paper addresses key challenges in microservice systems, including the difficulty of scheduling strategies adapting to complex dynamic environments, low resource utilization, and insufficient service quality assurance mechanisms. It proposes an intelligent scheduling and service quality optimization algorithm based on a multi-agent framework. In this method, microservice nodes are modeled as autonomous agents. Through collaborative learning and policy communication among agents, the system achieves a distributed perception of resource states and joint decision-making. The algorithm integrates Markov decision process modeling, policy gradient optimization, and composite reward function design. The scheduling actions cover three types of behaviors: scale-out, scale-in, and maintain. The method also considers multiple objective metrics, including task response latency, resource cost, and service completion rate. In the training process, a centralized training and distributed execution architecture is adopted. This enhances generalization in high-dimensional state spaces and improves policy stability. The method is evaluated through experiments involving various factors such as hyperparameters, data scale, observation dimensions, and environmental disturbances. These experiments comprehensively assess the adaptability and scheduling performance of the proposed approach in different typical scenarios. The results show that the method outperforms mainstream baselines in scheduling efficiency, service success rate, and resource utilization. It demonstrates strong robustness and overall performance advantages, effectively supporting the stable operation of microservice systems under high load and heterogeneous environments.