Deep Reinforcement Learning Oriented Queuing And Swarm Intelligence Framework For Adaptive Task Scheduling In End Edge Cloud Computing Environments

Abstract

The unprecedented growth of data intensive applications, cyber physical systems, and latency sensitive services has fundamentally reshaped the operational expectations of cloud, edge, and fog computing ecosystems. Modern computing environments are no longer limited to static cloud data centers but have evolved into complex end edge cloud architectures in which computation is dynamically distributed across geographically dispersed and resource heterogeneous nodes. Within this paradigm, task scheduling emerges as a central and persistent challenge, as the allocation of computational workloads to available resources directly determines system throughput, energy efficiency, delay, fairness, and economic sustainability. Traditional heuristic and metaheuristic approaches, while historically successful, increasingly struggle to cope with the stochasticity, scale, and heterogeneity that define contemporary distributed computing infrastructures. Consequently, there is a growing scholarly consensus that adaptive learning-based scheduling mechanisms are required to enable real time decision making under uncertainty.

This study develops and theorizes a novel reinforcement learning driven queuing and optimization framework for dynamic task scheduling across end edge cloud systems. Drawing upon foundational work in reinforcement learning theory, evolutionary computation, and queuing based performance modeling, the research constructs an integrated conceptual architecture that fuses deep Q learning with optimal queuing dynamics and swarm intelligence. In particular, the scheduling logic is informed by the theoretical insights of Kanikanti et al. (2025), who demonstrate that deep Q learning combined with optimal queuing theory can substantially improve dynamic task allocation in cloud computing environments. Their work provides a critical empirical and theoretical anchor for this study, which extends the paradigm into a broader end edge cloud orchestration context.

The article systematically analyzes how reinforcement learning agents can learn optimal task placement policies by observing system states such as queue length, processing capacity, latency, and energy consumption. These agents are embedded within a hybrid optimization framework that incorporates swarm-based search processes, genetic adaptation, and market-oriented scheduling strategies. The proposed model is not presented as a mathematical algorithmic artifact but as a deeply theorized scheduling ecosystem grounded in interdisciplinary research traditions. By integrating queuing theory with deep reinforcement learning, the framework accounts for both the temporal structure of task arrivals and the strategic adaptation of scheduling policies over time.

The results of this study, interpreted through extensive comparative reasoning against the existing literature, indicate that learning driven queuing-based schedulers can outperform static and metaheuristic approaches in terms of adaptability, delay minimization, and system robustness. Furthermore, the discussion situates these findings within broader debates about the future of intelligent computing infrastructures, arguing that end edge cloud systems represent not merely a technological shift but a paradigmatic transformation in how computational resources are conceptualized and governed. The article concludes by identifying key theoretical and practical implications for next generation scheduling systems and outlining a future research agenda centered on autonomous, self-optimizing distributed intelligence.