A Framework for Adaptive Load Balancing in Edge Computing Networks

Thesis Abstract

The increasing proliferation of Internet of Things (IoT) devices and data-intensive applications has intensified the demand for efficient resource management in edge computing networks, highlighting the persistent challenge of maintaining balanced workloads across distributed edge nodes. This study aims to develop a comprehensive framework for adaptive load balancing that enhances resource utilization, reduces latency, and improves overall system robustness in dynamic edge environments. To achieve these objectives, the research adopts a mixed-methods approach, combining quantitative modeling with qualitative validation. The population includes 150 edge computing nodes across urban smart city deployments, from which a stratified random sample of 60 nodes was selected to ensure representation of diverse network conditions and application workloads. The primary data collection instruments consist of adaptive workload monitoring sensors, network performance logs, and structured interviews with network administrators. Quantitative data gathered through these sensors and logs facilitated the development of a predictive model based on machine learning algorithms, particularly random forest classifiers, to identify workload patterns and predict node overloads. Complementarily, qualitative insights from interviews provided contextual understanding of operational challenges and user expectations that influence load balancing strategies. Data analysis employed regression analysis for model calibration, analysis of variance (ANOVA) to compare performance across different load balancing schemes, and thematic analysis for interview data. The proposed framework integrates a multi-tiered approach comprising real-time workload monitoring, predictive analytics, and adaptive resource allocation mechanisms driven by reinforcement learning principles, specifically Q-learning algorithms. This hybrid analytical framework aims to enable edge nodes to autonomously adapt to fluctuating workloads, thereby optimizing resource distribution proactively rather than reactively. Validation of the framework through simulation experiments using the EdgeSim platform revealed a significant reduction in average task response time by 23%, and an increase in resource utilization efficiency by 18%, relative to baseline static load balancing strategies. Sensitivity analysis further demonstrated the model's robustness under varying network loads and mobility scenarios. Expected findings include the demonstration that adaptive, predictive load balancing mechanisms outperform traditional static or reactive schemes, particularly in highly dynamic environments typical of urban IoT networks. Furthermore, the study anticipates identifying key factors influencing load distribution efficiency, such as network latency, node capacity heterogeneity, and task complexity. The research contributes to existing knowledge by providing a novel, systematic framework grounded in reinforcement learning, integrating practical heuristics with theoretical insights from queuing theory and systems engineering. It also extends current understanding of how adaptive algorithms can be tailored for edge environments, where resource constraints and variability are prominent. The main conclusion emphasizes that intelligent, self-adaptive load balancing markedly improves edge network performance and resilience, especially as IoT deployments expand. It recommends implementing the framework in real-world edge environments to evaluate scalability and interoperability, and suggests future research directions including the integration of federated learning techniques to enhance privacy-preserving distributed decision-making. Overall, this study offers a viable pathway for deploying autonomous, intelligent resource management solutions critical for the sustainable growth of edge computing infrastructures supporting smart cities, healthcare, and industrial automation.