Tag only storage for capacity optimised last level cache in chip multiprocessors

Surajit Das, Shirshendu Das, H. Kapoor
{"title":"Tag only storage for capacity optimised last level cache in chip multiprocessors","authors":"Surajit Das, Shirshendu Das, H. Kapoor","doi":"10.1109/ISVDAT.2016.8064886","DOIUrl":null,"url":null,"abstract":"Last Level Cache (LLC) plays an important role in increasing overall system performance of chip multiprocessor (CMP). The cores in a CMP has a private cache and a shared LLC whose size increases with each process generation. It has been observed that for different application on average 24% of the blocks are privately owned by the cores, and that the copies of these blocks in the LLC are redundant. This suggests that one needs to maintain only the tag portion of such private blocks in the LLC. Existing research has attempted to do this by partitioning the cache into tag-only and tag+data storage sections. This reduces the overall area of the LLC, but can degrade the performance to some extent. The tag-only storage stores the tags of the private blocks and if there are more such blocks in a set then these occupy the ways from the tag+data part. Thus, the chances of tag+data blocks staying in the cache reduces, affecting the performance. In this paper we propose to rectify the performance degradation by allowing more storage for the tag-only as well as tag+data parts. This is done by dynamically increasing the associativity of the sets. Experimental evaluation shows significant performance improvement with comparable area savings.","PeriodicalId":301815,"journal":{"name":"2016 20th International Symposium on VLSI Design and Test (VDAT)","volume":"75 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2016-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2016 20th International Symposium on VLSI Design and Test (VDAT)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ISVDAT.2016.8064886","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

Abstract

Last Level Cache (LLC) plays an important role in increasing overall system performance of chip multiprocessor (CMP). The cores in a CMP has a private cache and a shared LLC whose size increases with each process generation. It has been observed that for different application on average 24% of the blocks are privately owned by the cores, and that the copies of these blocks in the LLC are redundant. This suggests that one needs to maintain only the tag portion of such private blocks in the LLC. Existing research has attempted to do this by partitioning the cache into tag-only and tag+data storage sections. This reduces the overall area of the LLC, but can degrade the performance to some extent. The tag-only storage stores the tags of the private blocks and if there are more such blocks in a set then these occupy the ways from the tag+data part. Thus, the chances of tag+data blocks staying in the cache reduces, affecting the performance. In this paper we propose to rectify the performance degradation by allowing more storage for the tag-only as well as tag+data parts. This is done by dynamically increasing the associativity of the sets. Experimental evaluation shows significant performance improvement with comparable area savings.
标签仅存储容量优化的最后一级缓存在芯片多处理器
最后级缓存(Last Level Cache, LLC)对于提高芯片多处理器(CMP)的整体系统性能起着重要作用。CMP中的核心有一个私有缓存和一个共享LLC,其大小随着每个进程的生成而增加。我们观察到,对于不同的应用,平均24%的区块是由核心私有的,并且这些区块在LLC中的副本是冗余的。这表明只需要在LLC中维护这些私有块的标记部分。现有的研究试图通过将缓存划分为仅标记和标记+数据存储部分来实现这一点。这减少了LLC的总体面积,但在一定程度上降低了性能。仅标记存储存储私有块的标记,如果在一个集合中有更多这样的块,那么这些块将占用标记+数据部分的方式。因此,标记+数据块留在缓存中的机会减少了,从而影响了性能。在本文中,我们建议通过允许更多的存储仅标签和标签+数据部分来纠正性能下降。这是通过动态增加集合的结合性来实现的。实验评估显示了显著的性能改进和相当的面积节约。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信