Dynamic memory hierarchy performance optimization

Journal ArticleAlthough microprocessor performance continues to increase at a rapid pace, the growing processor-memory speed gap threatens to limit future performance gains. In this paper, we propose a novel configurable cache and TLB as an alternative to conventional two-level hierarchies. This organization leverages repeater insertion to provide low-cost configurability of size and speed. A novel configuration management algorithm dynamically measures hit and miss intolerance over intervals of instruction execution in order to tailor the cache and TLB organizations on-the-fly to improve memory hierarchy performance. The result is an average 14% improvement in IPC and a speedup of up to 1.55 across a broad class of applications compared to a conventional two-level hierarchy of identical total size