Dynamic Fair Cache Partitioning for Chip Multiprocessor

Abstract

Fairness is a critical issue because of some serious problems, such as thread starvation and priority inversion, it can arise and render the Operating System (OS) scheduler ineffective if no fair cache sharing which provided by the hardware. In order to improve the fairness of shared cache between threads in a chip multiprocessor, a dynamic fair partitioning policy of shared cache is proposed in this paper. There are threads sharing cache and simulating that it runs alone with a dedicated cache. At the end of each time interval T, policy evaluates the metrics of threads on shared and dedicated cache, then repartition cache to improve the fairness. Experimental data show that the dynamic fair partitioning policy of shared cache compared with LRU or other partitioning of shared cache can effectively enhance fairness and throughput has slightly increased for chip multiprocessor.