AML-TSCRA: A Machine Learning-Based Task Scheduling and Computational Resource Adaptive Allocation Method for Library Distributed Databases

INTRODUCTION: As libraries increasingly rely on distributed database systems to manage growing data volumes and service demands, conventional task scheduling and resource allocation strategies often struggle to adapt to dynamic workloads, leading to suboptimal resource utilization and performance bottlenecks. OBJECTIVES: This paper aims to develop an intelligent adaptive framework that jointly optimizes task scheduling and computational resource allocation, thereby improving responsiveness, load balancing, and overall system efficiency. METHODS: We propose AML-TSCRA (Adaptive Machine Learning-based Task Scheduling and Computational Resource Allocation), which integrates LSTM-based load forecasting with reinforcement learning-based scheduling and adaptive resource allocation based on real-time system states. RESULTS: Experiments on three representative datasets, including Amazon EC2 Logs, Google Cluster Data, and SLURM Workload Manager, show that AML-TSCRA achieves its strongest performance on Amazon EC2 Logs, with a task completion rate of 88.9%, resource utilization of 81.8%, an average response time of 1.31 s, and system throughput of 23.7 tasks/s. The model also shows moderate gains on Google Cluster Data and smaller but consistent advantages on the more heterogeneous SLURM workload. In addition, it demonstrates relatively better robustness than the learning-based baselines under noisy conditions. CONCLUSION: AML-TSCRA improves the efficiency and adaptability of distributed library-oriented service systems by combining predictive foresight, adaptive scheduling, and coordinated resource allocation. The proposed framework provides a practical basis for intelligent resource orchestration in modern data-intensive environments.