A Novel Method to Minimize Side Effects by Cache Contention on the Inclusive Multi-Core Shared Cache

摘要

Cache contention is annoying issue of the multi-core processors with private L1 cache and inclusive shared L2 cache. If core P1 runs with small working set which fits in its L1 cache while core P2 runs cache consuming process, loaded blocks in the shared L2 cache by P1 are extinguished by frequent access from P2. Furthermore, it incurs an inevitable invalidate to the L1 cache of the core P1 by the multi-level cache inclusion property. This study focused on this problem, and solves by access grouping with extension of cache status. Simulation results show the proposed method minimizes the side effects by the cache contention on the inclusive shared cache.