A simple statistical cache sharing model for multicores
Andreas Sandberg, David Black-Schaffer, Erik Hägersten · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2011
The introduction of multicores has made analysis of shared resources, such as shared caches and shared DRAM bandwidth, an important topic to study. We present two simple, but accurate, cache sharing models that use high-level data that can easily be measured on existing systems. We evaluate our model using a simulated multicore processor with four cores and a shared L2 cache. Our evaluation shows that we can predict average sharing in groups of four benchmarks with an average error smaller than 0.79% for random caches and 1.34% for LRU caches.