Abstract: Most processor adopts the inclusive cache hierarchy, which requires that the contents of the upper level cache are resident in last level cache. In-depth tests show that the Last-level Cache(LLC) block replacement algorithm has great performance impact on memory system performance. This paper proposes a new LLC cache replacement algorithm, called PLI, to select the best Least Recently Used(LRU) replacement block. Simulation results show that PLI has average 7% performance speedup over traditional approach.

Key words: cache replacement algorithm, Least Recently Used(LRU) algorithm, PLI algorithm, multi-level cache, Last-level Cache(LLC), inclusive-cache

摘要： 多数处理器中采用多级包含的cache存储层次，现有的末级cache块替换算法带来的性能开销较大。针对该问题，提出一种优化的末级cache块替换算法PLI，在选择丢弃块时考虑其在上级cache的访问频率，以较小的代价选出最优的LLC替换块。在时钟精确模拟器上的评测结果表明，该算法较原算法性能平均提升7%。

关键词: cache替换算法, LRU算法, PLI算法, 多级cache, 末级cache, 包容cache

CLC Number: 

• TP311