Modeling and optimizing NUMA effects and prefetching with machine learning

Both NUMA thread/data placement and hardware prefetcher configuration have significant impacts on HPC performance. Optimizing both together leads to a large and complex design space that has previously been impractical to explore at runtime.