발간년도 : [2020]
| 논문정보 |
|
| 논문명(한글) |
[Vol.15, No.2] An Effective Last-Level Cache Replacement and Bypass Policy Using Hit Count |
|
| 논문투고자 |
Young-Il Cho |
|
| 논문내용 |
Caching techniques are fundamental for bridging the performance gap between components in a computer system, such as the performance of processor and memory. To get more performance out of the cache hierarchy, processors will rely on effective cache replacement policies. Variability in generational behavior of cache blocks is a key challenge for cache replacement policies that aim to identify less reused blocks as early and accurately as possible to maximize the cache efficiency. Existing processors use variants of LRU policy to determine replacement targets. But, there is a big gap between what LRU policy provides and what the optimal replacement policy provides by Belady's MIN. In this paper, we observed that the number of hits of a block is similar to the average number of hits when previously residing in the cache. This observation is used to improve the efficiency of the LLC through a low cost but effective replacement policy. With little overhead, the proposed policy is an effective replacement policy that associates a predicted hit count to a block and evicts the block to have fewer remaining hits in the future. Also, blocks that are predicted as non-reusable blocks are bypassed to improve the cache efficiency. The proposal provides a 5.8% miss reduction and a 4.8% performance improvement over baseline LRU. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
