Adaptive Resource Scheduling in Distributed Systems via Reinforcement Learning under Dynamic Competition
- Abstract
-
To address the challenges faced by distributed systems in dynamic resource contention environments, such as uncertain task arrival, significant node load fluctuations, and insufficient adaptive capabilities of traditional scheduling strategies, a reinforcement learning-driven adaptive scheduling method for distributed systems is proposed. This method starts with the dynamic coupling characteristics of task requirements, node states, and resource contention, modeling the scheduling process as a sequential decision-making problem oriented towards long-term reward optimization. By constructing a competition-aware state representation mechanism, a unified characterization of multi-source heterogeneous information in complex operating environments is achieved. Furthermore, a context-enhanced state encoder is designed to improve the model's ability to express temporal dependencies, global load changes, and potential competition patterns. Combined with a multi-objective reward optimization mechanism, it collaboratively constrains job completion efficiency, resource utilization, load distribution balance, and system throughput, enabling the scheduling strategy to achieve more stable and effective decisions in complex resource game processes. Results show that the proposed method maintains strong environmental adaptability and global optimization capabilities in dynamically changing distributed scenarios, providing a feasible path for distributed resource scheduling that balances efficiency, balance, and overall system service capabilities.
- Downloads
- Published
- 2025-07-01
- Issue
- Vol. 5 No. 7 (2025)
- Section
- Articles