A Quantitative Analysis about the Cache Set-Level Utilization

It has seen a decrease in the amount of cache per core in order to make space for more cores. The efficiency of cache space becomes an important factor. The set-level unbalanced utilization is analyzed under the fixed capacity when associativity is varied and extended by increasing the number of physical ways (referred as Direct Extension Method (DEM)). We evaluate 29 benchmarks from Spec Cpu2000/2006 in Simics, compare the set-level unbalanced utilization between different associativity quantitatively and explain the trade-off of selecting a proper associativity for the cache with a certain fixed capacity from the view of the set-level unbalanced utilization and workloads. Furthermore, we explain the inherent deficiency of DEM through the statistical analysis of three typical benchmarks' set-level unbalanced utilization.