AI-Driven Data Placement in Distributed Storage: A Reinforcement Learning Framework for Workload Orchestration and QoS Optimization

Abstract

Modern distributed storage systems are challenged by increasingly complex optimizations of data placement across heterogeneous resources, without compromising performance and quality of service. Traditional data placement strategies rely on static policies, which cannot adapt to dynamic workload patterns and continuously changing system conditions. This article presents a holistic reinforcement learning framework that makes intelligent data placement decisions in distributed storage. The framework implements a Deep Reinforcement Learning-based placement engine with state-of-the-art policy optimization, a workload characterization system using time-series analysis and anomaly detection, and a QoS-aware orchestration layer with real-time feedback mechanisms. Initial simulations and theoretical analysis indicate great improvements in access latency reduction, resource utilization, and QoS compliance compared to traditional placement strategies, while reducing operational costs substantially. These preliminary findings reveal promising effectiveness in proactive management of dynamic workloads across heterogeneous environments by reinforcement learning-based decision making to bridge critical gaps in existing storage management approaches, and provide a way forward toward more efficient and adaptive storage.