Genetic Improvement of Last Level Cache

With increasing reliance on multi-core parallel computing performance is evermore dominated by interprocessor data communication typically provided by last level cache (LLC) shared between CPUs. In an 8 core 3.6 GHz desktop using multiple local searches, the Magpie parameter tuning genetic improvement (GI) system was able to reduce L3 cache access (load + stores) four fold on an existing open source 7000 line C PARSEC parallel computing VIPS image benchmark.