Optimal Memory Time Partitioning: Improve the Overall Performance of Concurrently Executing Applications In Chip-Multiprocessors
It is a problem to allocate cache capacity efficiently between concurrently executing applications in Chip-Multiprocessor (CMP). Partitioning is commonly used to solve this problem. Most of the existing partitioning schemes is to reduce the miss rate of the concurrently executing applications, but the optimal miss rate does not mean the best performance. This paper presents Optimal Memory Time cache Partitioning (OMTP), a lowoverhead, runtime mechanism that partitions a shared cache between multiple applications depending on the reduction in momery access time that each application is likely to obtain for a given amount of cache resources. The proposed mechanism monitors the actual penalty of all cache misses and use Auxiliary Tag Directory (ATD) to dynamically estimate the cache miss rate of executing applications. The experiment results show that OMTP has higher miss rate compared with utility-based cache partitioning (UCP), but it has better IPC throughput and weighted speedup.
Chip-Multiprocessor shared cache cache partitioning optimal memory time partitioning
Li Hao Xie Lunguo
Department of Computer Science National University of Defense Technology Changsha, China
国际会议
海口
英文
275-280
2011-02-22（万方平台首次上网日期，不代表论文的发表时间）