Multi-Objective Reinforcement Learning for Anticipatory Data Placement across Diverse Storage Technologies
Frontiers in Artificial Intelligence Research · 13.08.2025 · 2 (2) · 196-207
DOI: 10.71465/fair320 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Artificial Intelligence Research, online erschienen am 13.08.2025. Verfasst von Søren Jensen und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Modern data centers deploy diverse storage technologies including solid-state drives, persistent memory, optical storage, and tape systems to optimize cost-performance trade-offs across varying workload requirements. Traditional data placement strategies fail to effectively leverage the heterogeneous characteristics of diverse storage technologies, resulting in suboptimal resource utilization and missed opportunities for performance optimization. The challenge lies in anticipating future data access patterns while simultaneously optimizing multiple conflicting objectives including access latency, storage costs, energy consumption, and data durability across heterogeneous storage infrastructures. This study proposes a Multi-Objective Reinforcement Learning (MORL) framework for anticipatory data placement across diverse storage technologies. The framework employs Pareto-based optimization techniques combined with Deep Deterministic Policy Gradient (DDPG) algorithms to learn optimal placement policies that balance competing objectives. Predictive models forecast data access patterns and technology-specific performance characteristics, enabling proactive placement decisions that anticipate future system requirements. Experimental evaluation using real-world datacenter workloads demonstrates that the proposed framework achieves 47% reduction in average access latency while decreasing overall storage costs by 38% compared to traditional placement methods. The anticipatory approach reduces data migration overhead by 34% through proactive placement decisions, while the multi-objective optimization ensures balanced performance across all optimization criteria including energy efficiency and data durability requirements..
Zitieren
APA
Jensen, S., Hansen, M., & Kristensen, L. (2025). Multi-Objective Reinforcement Learning for Anticipatory Data Placement across Diverse Storage Technologies. Frontiers in Artificial Intelligence Research, 2(2), 196-207. https://doi.org/10.71465/fair320
Vancouver
Jensen S, Hansen M, Kristensen L. Multi-Objective Reinforcement Learning for Anticipatory Data Placement across Diverse Storage Technologies. Frontiers in Artificial Intelligence Research. 2025;2(2):196-207. doi:10.71465/fair320
BibTeX
@article{Jensen2025,
author = {Jensen, Søren and Hansen, Mette and Kristensen, Lars},
title = {Multi-Objective Reinforcement Learning for Anticipatory Data Placement across Diverse Storage Technologies},
journal = {Frontiers in Artificial Intelligence Research},
year = {2025},
volume = {2},
number = {2},
pages = {196--207},
doi = {10.71465/fair320}
}
EUVIMED