一种高效、可扩展细粒度缓存管理混合存储研究

计算机科学 ›› 2013, Vol. 40 ›› Issue (8): 79-82.

一种高效、可扩展细粒度缓存管理混合存储研究

姜国松

中国传媒大学信息与通信工程博士后流动站北京100024

出版日期:2018-11-16 发布日期:2018-11-16
基金资助:
本文受湖北省自然科学基金(2011CDC078)资助

Research about Efficient and Scalable Hybrid Memories at Fine-granularity Cache Management

JIANG Guo-song

Online:2018-11-16 Published:2018-11-16

摘要/Abstract

摘要： 混合主存储器由DRAM构成,它可用作cache来扩展非易失性存储器,相比传统的主存储器能够提供更大的存储能力。不过,要使混合存储器具有高性能和可扩展性,一个关键的挑战在于需要对缓存在DRAM中的数据的元数据(如标签)以一个细粒度的方式进行有效管理。基于这样的观察:利用DRAM缓存行的局部性,将元数据与元数据对应的数据存储在片外缓存中相同的行,使用一个小的缓冲区来只缓存最近被访问的片内缓存行,以降低细粒度DRAM缓存的开销。利用这种细粒度的DRAM高速缓存的灵活性和效率,还开发了一种自适应的策略来选择在数据迁移到DRAM时最佳的迁移粒度。在搭配了512MB的DRAM缓存的混合型存储系统中,建议使用8kB的片上缓存,这样,相比一个传统的8MB的SRAM元数据存储,即使没有考虑大的SRAM元数据存储的能源开销,也可以提升6％以内的性能,以及18％的能效节约。

关键词: 缓存,标签存储,非易失性存储器,混合主存储器

Abstract: Hybrid main memories are composed of DRAM which can provide much larger storage capacity than traditional main memories used as a cache to scalable non-volatile memories,such as phase-change memory．However,for hybrid main memories with high performance and scalability,a key challenge is to effectively manage the metadata (e.g.,tags) for data cached in DRAM in a fine- granularity．Based on this observation:storing metadata on off-chip cache line in the same row as their data corresponding to the metadata exploits DRAM row buffer locality,this paper reduced the overhead of fine-grained DRAM cache by using a small buffer to cache chip cache line which has recently been accessed．We also developed an adaptive policy to choose the best granularity when migrating data into DRAM．On a hybrid me-mory with a 512MB DRAM cache,our proposal using an 8KB on-chip buffer can increase the performance within 6％ and save 18％ better energy efficiency than a conventional 8MB SRAM metadata store,even when the energy overhead due to large SRAM metadata storage is not considered.

Key words: Cache,Tag memory,Non-volatile memories,Hybrid main memories

姜国松. 一种高效、可扩展细粒度缓存管理混合存储研究[J]. 计算机科学, 2013, 40(8): 79-82. https://doi.org/

JIANG Guo-song. Research about Efficient and Scalable Hybrid Memories at Fine-granularity Cache Management[J]. Computer Science, 2013, 40(8): 79-82. https://doi.org/

参考文献

[1] Lee B C,Ipek E,Mutlu O,et al.Architecting phase changememory as a scalable DRAM alternative[C]∥Proceedings of ISCA ’09．2009:171-182
[2] Qureshi M K,Srinivasan V,Rivers J A．Scalable high perfor-mance main memory system using phase-change memory techno-logy[C]∥Proceedings of ISCA’09．2009:101-112
[3] Dong X,Xie Y,Muralimanohar N,et al.Simple but effectiveheterogeneous main memory with on-chip memory controller support[C]∥Proceedings of SC’10．2010:773-778
[4] Zhao L,Iyer R,Illikkal R,et al.Exploring DRAM cache architectures for CMP server platforms[C]∥Proceedings of ICCD’07．2007:253-259
[5] Loh G,Hill M D．Efficiently enabling conventional block sizes for very large die-stacked DRAM caches[C]∥Proceedings of MICRO’11．2011:123-128
[6] Rixner S,Dally W J,Kapasi U J,et al．Memory access scheduling[C]∥Proceedings of ISCA’00．2000:58-69
[7] Zuravleff W K,Robinson T．Controller for a synchronous DRAM that maximizes throughput by allowing memory requests and commands to be issued out of order[P]．U.S．patent 5630096．1997
[8] Eyerman S,Eeckhout L．System-level performance metrics for multiprogram workloads[C]∥Proceedings of MICRO’08．2008:147-158
[9] Qureshi M K,Lynch D N,Mutlu O,et al.A case for MLP-aware cache replacement[C]∥Proceedings of ISCA’06．2006:211-221
[10] Jiang X,Madan N,Zhao L,et al.CHOP:Adaptive filter-basedDRAM caching for CMP server platforms[C]∥Proceedings of HPCA’10．2010:165-175
[11] Liptay J．Structural aspects of the System/360Model 85,II:The cache[J]．IBM Syst．J.,1968,7(1):15-21
[12] Seznec A．Decoupled sectored caches:conciliating low tag implementation cost and low miss ratio[C]∥Proceedings of ISCA’94．1994:19-28
[13] Wang H,Sun T,Yang Q．CAT-caching address tags-a technique for reducing area cost of on-chip caches[C]∥Proceedings of ISCA’95．1995:188-196
[14] Inoue K,Kai K,Murakami K．Dynamically variable linesizecache exploiting high on-chip memory bandwidth of merged DRAM/logic LSIs[C]∥Proceedings of HPCA’99．1999:119-128
[15] Johnson T L,Hwu W-M W．Run-TIRDe adaptive cache hierarchy management via reference analysis[C]∥Proceedings of ISCA’97．1997:134-145
[16] Mandelman J A,Dennard R H,Bronner G B,et al．Challengesand future directions for the scaling of dynamic random-access memory (DRAM)[J]．IBM J．Res．Dev.,2002,46(2/3):187-212

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed