HPC STRATEGIES FOR RADAR SIGNAL PROCESSING WORKLOADS USING DIVISIBLE LOAD FRAMEWORK

Abstract

SAR image processing involves extracting information from SAR images, enhancing their quality, and improving interpretability. Various algorithms and data processing techniques are used, ranging from simple image manipulation to complex pattern recognition algorithms. Distributed computing infrastructure can efficiently process SAR images, providing rapid results, reducing costs, and improving system efficiency. Efficient load distribution strategies are necessary for processing SAR image data on these computing clusters, tailored to the hardware resources and capable of scaling to handle changing workloads. The thesis focuses on applying Multi-instalment scheduling (MIS) methodologies to SAR image processing using the Divisible Load framework; for efficiently dividing the workload into smaller fractions for parallel execution, reducing processing time and improving performance. Two load distribution strategies are proposed: Multi-Installment Scheduling with Results Retrieval (MIS-RR) considers communication overheads and results retrieval, while APMIS-RR utilizes periodic internal installments and an adaptive last installment to optimize scheduling. Simulation frameworks and experiments validate the effectiveness of these strategies in reducing processing time, improving efficiency, and producing high-quality results in SAR image reconstruction applications.