An Adaptive Framework for Managing Heterogeneous Many-Core - ece uprm 2025

The dissertation presents an adaptive framework for managing heterogeneous many-core clusters, focusing on high-performance computing (HPC) systems that utilize specialized asymmetric multicores and computational accelerators like GPUs. It addresses challenges in programming models, resource management, and scheduling for data-intensive applications. The proposed framework includes an extended MapReduce programming model, efficient data prefetching techniques, and energy-aware resource scheduling mechanisms. Additionally, it explores advanced software engineering approaches using mixin-layers to enhance code reusability across different accelerator architectures. The evaluation demonstrates significant performance improvements and energy savings compared to traditional methods.