高阶路由器结构研究综述

摘要/Abstract

摘要：

随着高性能网络规模的增加，高阶路由器结构设计成为高性能计算中研究的重点和热点。使用高阶路由器，网络能实现更低的报文传输延迟、网络构建成本和网络功耗，同时高阶路由器的应用还可以提高网络可靠性。过去十年是高阶路由器发展最快的时期，对近年高阶路由器的研究进行了综述，并对未来发展趋势进行了预测，主要介绍了以YARC为代表的经典结构化设计以及“network within a network”等近年来涌现的新型设计方法。未来的研究重点是解决高阶路由器结构设计中遇到的缓存和仲裁等各种问题，并利用光互连等技术设计性能更好的结构。

关键词: 高性能计算, 高阶路由器, 结构化设计, 光互连

Abstract:

As the scale of high performance network increases, the design of high-radix router architecture is becoming a focus and a hotspot in the field of high performance computing. Using high-radix routers, the network can gain lower transmission latency, lower cost of network construction and lower power, and improve network reliability simultaneously. Over the past decade, high-radix routers have developed quickly. We present related research on high-radix routers and predict future development trends, mainly introducing the structured design represented by YARC and the “network within a network” approach. Our future study focuses on solving a number of problems such as buffer and arbitration, and we use some new techniques such as optical interconnection to design better architectures.Key words:high performance computing;high-radix router;structured design;optical interconnection

Key words: high performance computing, highradix router, structured design, optical interconnection

杨文祥，董德尊，雷斐，李存禄，吴际，孙凯旋. 高阶路由器结构研究综述[J]. 计算机工程与科学.

YANG Wen-xiang,DONG De-zun,LEI Fei,LI Cun-lu,WU Ji,SUN Kai-xuan. A survey on high-radix router architecture [J]. Computer Engineering & Science.

参考文献

［1］Dally W J,Towles B P. Principles and practices of interconnection networks［M］. New York:Elsevier,2004.
［2］Kim J. Highradix interconnection networks［D］. Palo Alto:Stanford University,2008.
［3］Satpathy S,Das R,Dreslinski R,et al. High radix selfarbitrating switch fabric with multiple arbitration schemes and quality of service［C］∥Proc of the 49th ACM Annual Design Automation Conference,2012: 406411.
［4］Kim J,Dally W J,Abts D. Flattened butterfly: a costefficient topology for highradix networks［J］. ACM SIGARCH Computer Architecture News,2007,35(2): 126137.
［5］Scott S,Abts D,Kim J,et al. The blackwidow highradix clos network［J］. ACM SIGARCH Computer Architecture News,2006,34(2): 1628.
［6］Kim J,Dally W J,Towles B,et al. Microarchitecture of a highradix router［J］. ACM SIGARCH Computer Architecture News,2005,33(2): 420431.
［7］Alverson R,Roweth D,Kaplan L. The gemini system interconnect［C］∥Proc of 2010 18th IEEE Symposium on High Performance Interconnects,2010: 8387.
［8］Faanes G, Bataineh A, Roweth D, et al. Cray cascade: A scalable HPC system based on a Dragonfly network［C］∥Proc of the International Conference on High Performance Computing,Networking,Storage and Analysis,2012: 103.
［9］Yang X J,Liao X K,Lu K,et al. The TianHe1A supercomputer: Its hardware and software［J］. Journal of Computer Science and Technology,2011,26(3): 344351.
［10］Liao X K,Pang Z B,Wang K F,et al. High performance interconnect network for Tianhe system［J］. Journal of Computer Science and Technology,2015,30(2): 259272.
［11］Chrysos N,Katevenis M. Scheduling in nonblocking buffered threestage switching fabrics［C］∥Proc of INFOCOM,2006: 113.
［12］Kim J,Dally W,Dally J,et al. Adaptive routing in highradix clos network［C］∥Proc of the ACM/IEEE SC 2006 Conference,2006: 7.
［13］Kao Y H,Alfaraj N,Yang M,et al. Design of highradix clos networkonchip［C］∥Proc of 2010 4th ACM/IEEE International Symposium on NetworksonChip(NOCS),2010: 181188.
［14］Fang M,Chen S,Wang K. Achieving high throughput in highradix switch［C］∥Proc of IEEE 10th International Conference on Trust, Security and Privacy in Computing and Communications(TrustCom),2011:14521456.
［15］Jeloka S,Das R,Dreslinski R G,et al. HiRise: A highradix switch for 3D integration with singlecycle arbitration［C］∥Proc of 2014 47th Annual IEEE/ACM International Symposium on Microarchitecture(MICRO),2014: 471483.
［16］Motoyoshi M. Throughsilicon via (TSV)［J］. Proceedings of the IEEE,2009,97(1): 4348.
［17］Sewell K,Dreslinski R G,Manville T,et al. Swizzleswitch networks for manycore systems［J］. IEEE Journal on Emerging and Selected Topics in Circuits and Systems,2012,2(2): 278294.
［18］Passas G,Katevenis M,Pnevmatikatos D. Crossbar NoCs are scalable beyond 100 nodes［J］. IEEE Transactions on ComputerAided Design of Integrated Circuits and Systems,2012,31(4): 573585.
［19］Ahn J H,Choo S,Kim J. Network within a network approach to create a scalable highradix router microarchitecture［C］∥Proc of 2012 IEEE 18th International Symposium on High Performance Computer Architecture (HPCA),2012: 112.
［20］Ahn J H,Son Y H,Kim J. Scalable highradix router microarchitecture using a network switch organization［J］. ACM Transactions on Architecture and Code Optimization (TACO),2013,10(3): 17.
［21］Chrysos N,Minkenberg C,Rudquist M,et al. SCOC: Highradix switches made of bufferless clos networks［C］∥Proc of 2015 IEEE 21st International Symposium on High Performance Computer Architecture(HPCA),2015: 402414.
［22］Underwood K D,Borch E,Sizer J,et al. Evaluating ondie interconnects for a 4 TB/s router［C］∥Proc of the 27th International ACM Conference on International Conference on Supercomputing,2013: 203212.
［23］Grot B,Hestness J,Keckler S W,et al. Express cube topologies for onchip interconnects［C］∥Proc of IEEE 15th International Symposium on High Performance Computer Architecture,2009: 163174.
［24］Chao H J,Jing Z,Liew S Y. Matching algorithms for threestage bufferless Clos network switches［J］. Communications Magazine,2003,41(10): 4654.
［25］Knights A P. Silicon photonics［M］. Bristol: IOP Publishing,2008.
［26］Joshi A, Batten C, Kwon Y J, et al. Siliconphotonic clos networks for global onchip communication［C］∥Proc of the 2009 3rd ACM/IEEE International Symposium on NetworksonChip, 2009: 124133.
［27］Krishnamoorthy A V,Ho R,Zheng X,et al. Computer systems based on silicon photonic interconnects［J］. Proceedings of the IEEE,2009,97(7): 13371361.
［28］Rong H,Jones R,Liu A,et al. A continuouswave Raman silicon laser［J］. Nature,2005,433(7027): 725728.
［29］The 50 Gbps Si Photonics Link［EB/OL］.［20160213］.http:∥www.intel.com/.
［30］Binkert N,Davis A,Jouppi N P,et al. The role of optics in future high radix switch design［C］∥Proc of 2011 38th Annual International Symposium on Computer Architecture(ISCA),2011: 437447.
［31］Binkert N,Davis A,Jouppi N P,et al. Optical high radix switch design［J］. IEEE Micro,2012,32(3): 100109.
［32］Mora G,Flich J,Duato J,et al. Towards an efficient switch architecture for highradix switches［C］∥Proc of the 2006 ACM/IEEE Symposium on Architecture for Networking and Communications Systems,2006: 1120.
［33］Eberle H, Garcia P J, Flich J, et al. Highradix crossbar switches enabled by proximity communication［C］∥Proc of the 2008 ACM/IEEE Conference on Supercomputing,2008: 32.
［34］Kim J,Balfour J,Dally W. Flattened butterfly topology for onchip networks［C］∥Proc of the 40th Annual IEEE/ACM International Symposium on Microarchitecture,2007: 172182.
［35］Oki E,Jing Z,RojasCessa R,et al. Concurrent roundrobinbased dispatching schemes for Closnetwork switches［J］. IEEE/ACM Transactions on Networking,2002,10(6): 830844.
［36］Haney M,Nair R,Gu T. Chipscale integrated optical interconnects:A key enabler for future highperformance computing［C］∥Proc of SPIE—International Society for Optics and Photonics,2012: 82670X82670X12.
［37］Jiang N, Kim J, Dally W J. Indirect adaptive routing on large scale interconnection networks［J］. ACM SIGARCH Computer Architecture News,2009,37(3): 220231.
［38］Cray XC Series Network［EB/OL］.［20160213］.http:∥www.cray.com/.

[1]	吴铁彬, 过锋, 王谛. 面向E级计算的高性能处理器核心运算架构研究进展[J]. 计算机工程与科学, 2023, 45(05): 761-771.
[2]	陈奉贤. 基于NR-Transformer的集群作业运行时间预测[J]. 计算机工程与科学, 2022, 44(07): 1181-1190.
[3]	曹继军. 面向HPC和DC的可重构光互连网络体系结构综述[J]. 计算机工程与科学, 2022, 44(06): 951-963.
[4]	梁崇山, 戴艺, 徐炜遐. 一种基于Chiplet集成技术的超高阶路由器设计[J]. 计算机工程与科学, 2022, 44(02): 207-213.
[5]	袁国兴, 张云泉, 袁良. 2021年中国高性能计算机发展现状分析[J]. 计算机工程与科学, 2021, 43(12): 2091-2097.
[6]	袁远, 李世杰, 邢建英, 蒋句平. E级高性能计算机系统中监控分系统的挑战与设计[J]. 计算机工程与科学, 2021, 43(08): 1366-1375.
[7]	吴君楠, 欧洋, 李琰. 基于LAMP的高性能计算用户组织架构管理系统设计与实现[J]. 计算机工程与科学, 2021, 43(02): 235-241.
[8]	刘杰, 龚春叶, 杨博, 郭晓威, 甘新标, 李胜国, 李超, 陈旭光, 肖调杰, 穆利安, 宋敏, 赵冬勇, 鞠羽中. YH-ACT：热工流体力学并行应用程序[J]. 计算机工程与科学, 2021, 43(01): 58-69.
[9]	袁国兴, 张云泉, 袁良. 2020年中国高性能计算机发展现状分析[J]. 计算机工程与科学, 2020, 42(12): 2103-2108.
[10]	李哲, 谭郁松, 李宝, 余杰. 面向HPC的函数计算冷启动优化[J]. 计算机工程与科学, 2020, 42(11): 1973-1980.
[11]	李琼, 宋振龙, 袁远, 谢徐超. 一种基于NVMeoF存储池的分域共享并发存储架构[J]. 计算机工程与科学, 2020, 42(10高性能专刊): 1711-1719.
[12]	王耀华, 郭阳. 面向HPC的高性能微处理器研究进展[J]. 计算机工程与科学, 2020, 42(10高性能专刊): 1742-1748.
[13]	宋振龙, 李小芳, 李琼, 谢徐超, 魏登萍, 董勇, 王睿伯. BeeGFS并行文件系统性能优化技术研究[J]. 计算机工程与科学, 2020, 42(10高性能专刊): 1765-1773.
[14]	张昆, 张伟, 卢凯, 董勇, 戴屹钦. 百亿亿次级系统进程管理接口综述[J]. 计算机工程与科学, 2020, 42(10高性能专刊): 1774-1783.
[15]	赵奇奇. “魔方-3”高性能计算机运维管理平台设计与实现[J]. 计算机工程与科学, 2020, 42(10高性能专刊): 1807-1814.