期刊文献+

基于紧耦合单跳步多平面架构的高端服务器设计 被引量:5

Design of high-end server based on tightly-coupled single-hop multi-plane architecture
下载PDF
导出
摘要 针对高端服务器设计面临的可扩展性问题,提出了一种紧耦合单跳步多平面(TSMP)体系结构设计方法。该方法采用双侧多平面互连结构,支持8-32路规模无缝扩展;基于两级目录结构的高速缓存一致性实现方法,支持高并发一致性访问和高效冲突处理,有效降低一致性访问传输、处理延迟。该方法已应用于浪潮32路K1高端服务器的设计,对设计的系统进行了内存性能、处理性能和可扩展性测试,测试结果表明,采用该设计可使高端服务器的计算、访存性能随系统规模从单路到32路线性增长。K1高端服务器支持基于QPI1.0协议的Intel安腾(Itanium)4核CPU-Tukwila和8核CPU-Polson,是中国研制的首台投入商业化应用的高端服务器。 A new method based on the tightly-coupled single-hop multi-plane(TSMP)architecture is put forward to solve the system expansibility issue in high-end server design. It adopts a two-side single-hop multi-plane topology to sup- port the seamless extending of system from 8 to 32-way. A two-tier directory based cache coherence maintenance method is applied to sustain both high parallel cache coherence requests and high efficiency conflicts handling, which reduces the system transmission delay and disposal delay remarkably. The TSMP architecture has beea used in the design of an Inspur K1 high-end server. Testing results of Stream, SPEC CPU and high pertormance Linpack (HPL) indicate that K1 system performance increases linearly from 1 to 32-way. The K1 high-end server supports the QPI1.0 based Intel Itanium 4-core Tukwila and 8-core Poison CPUs. It is China' s first independent commercial high-end server.
出处 《高技术通讯》 CAS CSCD 北大核心 2014年第2期111-116,共6页 Chinese High Technology Letters
基金 863计划(2008AA01A202) 973计划(2010CB735905)资助项目
关键词 缓存一致性非均匀存储访问(CC-NUMA) 紧耦合单跳步多平面(TSMP) QPI协议 CACHE一致性 目录cache cache coherence non-uniform memory access( CC-NUMA), tightly-coupled single-hop multi-plane(TSMP), QPI, cache coherence, cache directory
  • 相关文献

参考文献8

  • 1Laudon J, Lenoski D. The SGI Origin : a ccNUMA highly scalable server. In: Proceedings of the ACM 24th Annual International Symposium on Computer Architecture, New York, USA, 1997. 241-251.
  • 2Gostin G, Collard J-F, Collins K. The architecture of HP superdome shared-memory muhiprocessor. In : Proceedings of the ACM 19th Annual International Conference on Su- percomputing, New York, USA ,2005. 239-245.
  • 3Aono F, Kimura M. The AZUSA 16-way Itanium server. IEEE Micro ,2000,20(5 ) :54-60.
  • 4Gharachorlooy K, Sharma M, Steely S, et al. Architecture and design of AlphaServer GS320. ACM Sigplan Notices, 2000,35( 11 ) :13-24.
  • 5Conway P, Hughes B. The AMD Opteron northbridge ar- chitecture. IEEE Micro ,2007,27 (2) : 10-21.
  • 6Feehrer J, Rotker P, Shih M, et al. Coherency hub design for muhisoeket SUN servers with coohhreads technology. IEEE Micro ,2009,29(4 ) :36-47.
  • 7Kota R, Oehler R. HORUS : Large-scale symmetric muhi- processing for Opteron system. IEEE Micro ,2005,25 ( 2 ) : 30-40.
  • 8Aeacio M. E, Gonzalez J, Garcia J. M, et al. A two-level directory architecture for highly sealable ce-NUMA muhi- processors. IEEE Transactions on Parallel and Distributed Systems,2005,16( 1 ) :67-79.

同被引文献20

引证文献5

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部