To address the issues of drastic load fluctuations, unstable service quality, resource supply imbalances, and increased operating costs in backend systems under high-concurrency business environments during resource scheduling, this paper proposes a dynamic resource allocation and performance optimization method based on Deep Q Learning for elastic control of backend systems. This method models the resource control process as a continuous interactive decision-making task, constructing a system state representation that includes information such as request intensity, computational resource consumption, task queuing status, response latency, and processing capacity to jointly characterize complex operational characteristics. Based on this, a discrete action space is established around control behaviors such as resource expansion, resource contraction, and maintaining the status quo. A reward mechanism that balances service assurance, processing efficiency, resource utilization, and control costs is designed to enhance the adaptability of policy learning to the actual operational goals of the backend system. Furthermore, by combining experience replay and target network update mechanisms, the stability of value function estimation and the consistency of the decision-making process are improved, enabling the model to learn more reasonable resource allocation strategies under dynamic load conditions. The results show that the proposed method can effectively improve the elastic control level of backend systems in high- concurrency scenarios, exhibiting good comprehensive performance in terms of service quality maintenance, throughput assurance, resource allocation efficiency, and overall operational cost control. This research provides a practically valuable methodology for intelligent resource management and performance optimization of backend systems.
pdf
