Research on a Deep Reinforcement Learning-Based Adaptive Replica Allocation Method for Data Replica Placement Optimization in Distributed Storage Systems

Authors
  • Bo Zhang

    The University of Chicago, Chicago, USA

  • Ilyar Abdurahman

    Stevens Institute of Technology, Hoboken, USA

Abstract
To address the challenges of data replica placement in distributed storage systems, such as significant fluctuations in access load, dynamic changes in node resource status, insufficient adaptability of traditional rule-based methods, and difficulties in coordinating multi-objective optimization, this paper proposes a deep reinforcement learning-based adaptive replica allocation method for replica placement optimization. This method models replica placement as a continuous decision-making process, comprehensively considering node storage utilization, request intensity, network operating status, fault risk information, and historical hotspot changes within a unified framework. It constructs a state representation for complex system environments and achieves coordinated optimization among replica quantity adjustment, placement location selection, and migration control through a policy learning mechanism. In terms of method design, this paper enhances the model's ability to perceive dynamic environments and optimize long-term returns through key steps such as state encoding, latent feature extraction, policy network decision-making, reward function construction, and parameter iterative updates. This enables the replica allocation process to balance access latency, system reliability, load balancing, and resource overhead control. To verify the effectiveness of the proposed method in distributed storage replica management scenarios, this paper selects publicly available cluster operation trajectory data to construct an experimental environment and compares it with related methods in the same direction. The results show that the proposed method outperforms the traditional method in terms of overall performance, achieving a more reasonable replica layout and more stable policy decisions under complex load conditions, demonstrating strong global coordination and environmental adaptability. This research indicates that introducing deep reinforcement learning into the replica placement optimization task of distributed storage systems can provide a new technical approach for ensuring data availability, optimizing resource scheduling, and improving service quality.
References
Cover Image
Downloads
Published
2025-08-30
Section
Articles
License
Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.

Academic Scholar Hub Press also operates under the Creative Commons Licence CC-BY 4.0. This allows for copy and redistribute the material in any medium or format for any purpose, even commercially. The premise is that you must provide appropriate citation information.

How to Cite

Zhang, B., & Abdurahman, I. (2025). Research on a Deep Reinforcement Learning-Based Adaptive Replica Allocation Method for Data Replica Placement Optimization in Distributed Storage Systems. Journal of Computer Technology and Software, 4(8). https://doi.org/10.5281/zenodo.22791573