In this paper, we present the Tianhe-2 interconnect network and message passing services. We describe the architecture of the router and network interface chips, and highlight a set of hardware and software features e...In this paper, we present the Tianhe-2 interconnect network and message passing services. We describe the architecture of the router and network interface chips, and highlight a set of hardware and software features effectively supporting high performance communications, ranging over remote direct memory access, collective optimization, hardwareenable reliable end-to-end communication, user-level message passing services, etc. Measured hardware performance results are also presented.展开更多
This paper presents an overview of TianHe-lA （TH-1A） supercomputer, which is built by National University of Defense Technology of China （NUDT）. TH-1A adopts a hybrid architecture by integrating CPUs and GPUs, and...This paper presents an overview of TianHe-lA （TH-1A） supercomputer, which is built by National University of Defense Technology of China （NUDT）. TH-1A adopts a hybrid architecture by integrating CPUs and GPUs, and its interconnect network is a proprietary high-speed communication network. The theoretical peak performance of TH-1A is 4700TFlops, and its LINPACK test result is 2566TFlops. It was ranked the No. 1 on the TOP500 List released in November, 2010. TH-1A is now deployed in National Supercomputer Center in Tianjin and provides high performance computing services. TH-1A has played an important role in many applications, such as oil exploration, weather forecast, bio-medical research.展开更多
集中研究了非结构化的数据存储和查询。为了在保证查询成功率的同时最小化总的能耗,分别在存储受限和不受限两种情况下,建立了MESQ(minimizing energy on success fulquery)优化问题模型,给出并证明了最优的复本和查询个数。在此基础上...集中研究了非结构化的数据存储和查询。为了在保证查询成功率的同时最小化总的能耗,分别在存储受限和不受限两种情况下,建立了MESQ(minimizing energy on success fulquery)优化问题模型,给出并证明了最优的复本和查询个数。在此基础上,还设计了一个实用的分布式数据分发算法:BubbleGeocast,其主要包含精确自适应快速分发和基于拒绝的均匀分发两个部分,其中前者用自适应分支的方法加速数据扩散,并精确控制总的复本个数;后者根据每个节点Voronoi单元面积,决定是否接受或拒绝这个报文。从而保证了复本和查询分发的精确性、实时性、均匀性、顽健性。最后,详细的理论分析和模拟实验进一步验证了其性能。分析和实验表明,同已有工作相比,在相同查询成功率时,BubbleGeocast能量有效性平均提高了约30%,复本分发的延迟平均缩短了约30%,成功查询的延迟平均缩短了约50%。展开更多
基金This work was partially supported by the National High Technology Research and Development 863 Program of China under Grant No. 2012AA01A301 and the National Natural Science Foundation of China under Grant No. 61120106005. Acknowledgements The Tianhe-2 project is a great team effort and benefits from the cooperation of many individuals at NUDT. We would like to thank the entire Tianhe-2 development, applications, and bench- marking teams, and all the people who have contributed to the system in a variety of ways.
文摘In this paper, we present the Tianhe-2 interconnect network and message passing services. We describe the architecture of the router and network interface chips, and highlight a set of hardware and software features effectively supporting high performance communications, ranging over remote direct memory access, collective optimization, hardwareenable reliable end-to-end communication, user-level message passing services, etc. Measured hardware performance results are also presented.
基金Supported by the National High Technology Research and Development 863 Program of China under Grant No. 2009AA01A128
文摘This paper presents an overview of TianHe-lA （TH-1A） supercomputer, which is built by National University of Defense Technology of China （NUDT）. TH-1A adopts a hybrid architecture by integrating CPUs and GPUs, and its interconnect network is a proprietary high-speed communication network. The theoretical peak performance of TH-1A is 4700TFlops, and its LINPACK test result is 2566TFlops. It was ranked the No. 1 on the TOP500 List released in November, 2010. TH-1A is now deployed in National Supercomputer Center in Tianjin and provides high performance computing services. TH-1A has played an important role in many applications, such as oil exploration, weather forecast, bio-medical research.
文摘集中研究了非结构化的数据存储和查询。为了在保证查询成功率的同时最小化总的能耗,分别在存储受限和不受限两种情况下,建立了MESQ(minimizing energy on success fulquery)优化问题模型,给出并证明了最优的复本和查询个数。在此基础上,还设计了一个实用的分布式数据分发算法:BubbleGeocast,其主要包含精确自适应快速分发和基于拒绝的均匀分发两个部分,其中前者用自适应分支的方法加速数据扩散,并精确控制总的复本个数;后者根据每个节点Voronoi单元面积,决定是否接受或拒绝这个报文。从而保证了复本和查询分发的精确性、实时性、均匀性、顽健性。最后,详细的理论分析和模拟实验进一步验证了其性能。分析和实验表明,同已有工作相比,在相同查询成功率时,BubbleGeocast能量有效性平均提高了约30%,复本分发的延迟平均缩短了约30%,成功查询的延迟平均缩短了约50%。
