Deep Reinforcement Learning Integrated Queuing Architectures For Adaptive Task Orchestration In Cloud Computing Environments

Abstract

The accelerating expansion of cloud computing infrastructures has transformed the contemporary digital economy by enabling scalable, elastic, and distributed access to computational resources across heterogeneous environments. As enterprise, scientific, and consumer workloads continue to grow in complexity and volume, the demand for intelligent task scheduling frameworks that can dynamically balance performance, efficiency, and reliability has become increasingly critical. Classical queuing theory has historically provided the foundational analytical backbone for understanding service systems, particularly within computing and telecommunication domains, yet the static and assumption-bound nature of many traditional queuing models has struggled to accommodate the stochastic volatility and multi-objective constraints inherent in modern cloud ecosystems (Kleinrock, 1975; Gross et al., 2008). Simultaneously, reinforcement learning has emerged as a powerful paradigm for decision-making under uncertainty, offering the capacity to adaptively learn optimal control strategies from environmental feedback rather than relying on predetermined rules. The convergence of these two intellectual traditions has opened a promising research frontier, in which learning-driven schedulers are informed and constrained by queuing-theoretic principles.