Reducing Cache Misses for CC-NUMA by Careful Page-mapping

Abstract

Careful page mapping has been shown in the past to be effective for reducing cache conflicts on both uniprocessor and Uniform Memory Access (UMA) multiprocessors. We extend previous page-mapping schemes to Cache-Coherent Non-Uniform Memory Access (CG-NUMA) multiprocessors. These extensions maintain the program's data-task affinity, which is important to CG-NUMA, while reducing cache set conflicts by carefully selecting the page frames. Using an execution-driven simulator that simulates a CC-NUMA machine with a 2-MB secondary cache and a 16-KB primary cache on each of the 16 four-issue super-scalar processors, we find that a simplistic application of page-coloring performs worse than bin-hopping by 10-45%, while by hashing the page color with part of the MID bits, page-coloring can perform closely to bin-hopping.