A Framework for Adaptive Resource Allocation in Edge Computing Environments

Thesis Abstract

The rapid proliferation of Internet of Things (IoT) devices and the increasing demand for low-latency processing have positioned edge computing as a pivotal paradigm in contemporary distributed systems. However, the dynamic and heterogeneous nature of edge environments presents significant challenges in resource management, often resulting in suboptimal utilization, increased latency, and diminished Quality of Service (QoS). This study aims to develop a comprehensive framework for adaptive resource allocation tailored to the unique operational context of edge computing environments, thereby enhancing system efficiency, scalability, and responsiveness. The primary objective of this research is to design, implement, and evaluate an adaptive resource management framework that intelligently allocates computational, storage, and network resources based on real-time data, workload demands, and contextual factors. To achieve this, the study sets out specific objectives (1) to analyze existing resource allocation strategies in edge computing; (2) to identify key parameters influencing resource demands; (3) to develop an adaptive model incorporating machine learning techniques for predictive resource management; and (4) to validate the proposed framework through simulation and real-world testing. The research adopts a mixed-methods approach, combining quantitative simulations with qualitative system evaluations. The population consists of edge nodes within a metropolitan IoT network, with a sample size of 50 geographically dispersed edge devices selected via stratified random sampling to ensure representation across different operational scales and functionalities. Data collection involves extracting system logs, performance metrics, and workload characteristics over a six-month period, supplemented by expert interviews to inform model development. Data analysis employs a combination of regression analysis to identify significant predictors of resource demands, clustering algorithms to segment workload patterns, and time-series forecasting using Long Short-Term Memory (LSTM) neural networks to predict future resource needs. The analytical framework is anchored within the Resource-Based View (RBV) theory to conceptualize how dynamic resource allocation can generate competitive advantage in edge systems. Additionally, the framework incorporates principles from the Self-Organizing Map (SOM) theory to facilitate adaptive decision-making processes. Key anticipated findings include the identification of crucial workload parameters that influence resource contention, the demonstration of the effectiveness of machine learning-based predictive models in optimizing resource distribution, and insights into the trade-offs between resource utilization and QoS metrics such as latency and throughput. It is expected that the proposed framework will outperform static allocation schemes, reducing average task response time by at least 25% and improving resource utilization rates by approximately 15%. This study significantly contributes to knowledge by integrating machine learning techniques into a cohesive, adaptable resource management framework specifically designed for edge environments, an area currently marked by fragmentation and limited automation. It offers a scalable and context-aware approach that can be tailored to diverse use cases, from smart cities to industrial automation. The primary conclusion emphasizes that adaptive, data-driven resource management enhances the operational efficiency of edge systems under varying workload conditions. Recommendations include the deployment of the framework in heterogeneous edge networks, continuous refinement of predictive models with live data, and exploration of integrating the framework with existing orchestration platforms. Future research should investigate the application of reinforcement learning algorithms within the framework to further improve decision-making processes, and extend validation through longitudinal real-world implementations across different sectors.