Degree: Doctor of Philosophy in Electrical & Computer Engineering
Committee Members: Dr. Samee Khan, Dr. Asad Malik, Dr. Bryan Jones, Dr. Yu Luo
Abstract:
The rapid proliferation of IoT devices has led to an exponential increase in data generation, necessitating efficient and scalable stream processing solutions within resource-constrained edge networks. This dissertation investigates the challenges of load balancing in distributed stream processing platforms, addressing the need for fairness, real-time processing, and resource optimization. We explore various approaches, including probabilistic modeling and machine learning-based solutions, for load balancing, and to enhance system performance under dynamic workloads.
Our research contributes to the literature by proposing novel frameworks that leverage Gaussian Process Regression (GPR) for performance modeling, enabling predictive analysis for optimizing resource allocation. Additionally, we introduce an intelligent data dissemination mechanism that utilizes machine learning for multi-hop routing in edge networks, reducing congestion and improving overall system efficiency. To further enhance resource utilization, we present iGenEdge, a federated edge computing framework designed for deploying computationally expensive models such as Large Language Models (LLMs) on resource-constrained devices. This framework dynamically allocates resources to meet fluctuating demands, ensuring improved service quality across IoT applications.
Furthermore, we benchmark the performance of distributed stream processing frameworks in both virtual and physical testbeds, utilizing Raspberry Pi clusters to assess real-world constraints. Our findings highlight significant performance variations between virtualized environments and physical edge networks, underscoring the need for adaptive resource management strategies tailored for constrained hardware.
This dissertation advances edge computing by addressing key challenges in real-time stream processing, load balancing, and resource optimization. Future work will focus on the integration of AI-driven decision-making models, adaptive data-priority mechanisms, and federated learning techniques to further enhance scalability and efficiency in large-scale IoT deployments. These contributions pave the way for more resilient and intelligent edge computing systems, capable of handling the complexities of modern data-driven applications in diverse environments.
