Intelligent Queue-Driven Task Scheduling Using Deep Reinforcement Learning In End–Edge–Cloud Environments

Abstract

The accelerating convergence of cloud computing, mobile edge infrastructures, and intelligent end devices has transformed how computational workloads are generated, distributed, and executed across modern digital ecosystems. This transformation has produced unprecedented heterogeneity in task characteristics, network conditions, and resource availability, thereby exposing the limitations of traditional static and heuristic based scheduling approaches. Recent scholarly discourse increasingly emphasizes the necessity of intelligent, learning driven scheduling mechanisms that can dynamically adapt to stochastic workloads and time varying system states in end edge cloud environments, a position strongly supported by recent advances in deep reinforcement learning and optimal queuing theory. Within this evolving landscape, the integration of deep Q learning with queue aware task scheduling has emerged as a promising paradigm for optimizing latency, throughput, and energy efficiency in cloud computing systems, particularly when workloads originate from delay sensitive Internet of Things and vehicular networks (Zhou et al., 2021; Jiang et al., 2022).

This article develops a comprehensive theoretical and methodological framework for deep reinforcement learning driven dynamic optimal task scheduling grounded in queue theoretic principles and orchestrated across end edge cloud architectures. Building upon the recent empirical and algorithmic contributions of Kanikanti et al. (2025), who demonstrated the viability of deep Q learning based optimal queuing for cloud task scheduling, this study extends their conceptual foundation into a broader systems oriented and analytically rigorous architecture that integrates multi layer orchestration, adaptive offloading, and predictive resource allocation. By synthesizing insights from reinforcement learning based scheduling in cloud computing, mobile edge computing, and vehicular networks, the research articulates a unifying perspective that situates queue aware deep Q learning as the core intelligence for distributed computation management (Li et al., 2018; Ning et al., 2019; Jazayeri et al., 2020).

The study is structured around a deeply elaborated methodological design that describes how state representations, reward functions, and policy updates are harmonized with queuing dynamics and end edge cloud orchestration. Instead of treating queues as passive buffers, they are reconceptualized as dynamic signals of congestion, delay, and service quality, which are continuously interpreted by a learning agent to inform scheduling decisions, a view that resonates with the delay oriented reinforcement learning paradigms proposed by Zhou et al. (2021). The proposed framework is theoretically validated through a rich descriptive and interpretive results section that maps expected performance behaviors to established findings in the literature on reinforcement learning driven offloading and cloud optimization (Chen et al., 2020; Asghari et al., 2020).

The discussion section situates these results within ongoing scholarly debates about the scalability, convergence, and interpretability of deep reinforcement learning in distributed computing systems, critically engaging with both optimistic and skeptical perspectives on learning based orchestration (Ren et al., 2019; Ren et al., 2020). Through extensive theoretical reasoning and cross study synthesis, the article demonstrates that queue aware deep Q learning offers not merely incremental improvements over classical schedulers but represents a paradigmatic shift toward self optimizing computational ecosystems.

By providing a publication ready, deeply contextualized, and theoretically grounded account of deep Q learning driven optimal queuing for task scheduling, this research contributes a durable conceptual foundation for future empirical investigations and practical deployments in intelligent cloud and edge computing systems.