Development of supercomputers in Japan: Hardware and software

Japanese supercomputer development activities in the industry is reviewed. Architectures and softwares of Japanese supercomputer systems in the last 20 years are described and discussed. Special emphasis is put on the similarities and differences of the systems.     

New trends revive supercomputing industry

Although grid computing - which links disparate machines so that they can function as a distributed supercomputer - has become an increasingly popular focus of high-performance-computing research, the traditional supercomputing industry has languished until recently. This occurred largely because the easing of the Cold War in the early 1990s reduced government use of and spending on supercomputer technology. However, the industry is now reviving because of the development of low-cost supercomputer clusters that use commodity chips. Clusters of commodity computers linked by highspeed interconnect technologies, such as InfiniBand, have put supercomputers within reach of new users. Meanwhile, longtime supercomputer vendors like Cray are making a comeback with systems based on vector units, powerful CPUs dedicated to floating-point and matrix calculations. Vector machines satisfy the demand by governments and large industries for supercomputers that can conduct complex tasks such as nuclear-weapons simulations, pharmaceutical drug modeling, mining of large data sets, and geological analysis to find oil deposits.     

一种基于消息总线的可用性管理技术

分析了大规模并行机中可用性管理设计方面存在的问题,提出了基于消息总线的可用性管理模式,介绍了消息总线的设计思路和方法,完成了对该总线的测试。结合一种并行机的具体情况,阐述了基于消息总线的可用性管理技术,并讨论了其设计与实现。     

大规模并行时域有限差分法电磁计算研究

基于我国超级计算机平台,开展了大规模并行时域有限差分法(Finite-Difference Time-DomainFDTD)的性能和应用研究。在我国首台百万亿次"魔方"超级计算机、具有国产CPU的"神威蓝光"超级计算机和当前排名世界第一的"天河二号"超级计算机上就并行FDTD方法的并行性能进行了测试,并分别突破了10000 CPU核,100000 CPU核和300000 CPU核的并行规模。在不同测试规模下,该算法的并行效率均达到了50%以上,表明了本文并行算法具有良好的可扩展性。通过仿真分析多个微带天线阵的辐射特性和某大型飞机的散射特性,表明本文方法可以在不同架构的超级计算机上对复杂电磁问题进行精确高效电磁仿真。     

Ballistic Missile Defense: A Supercomputer Challenge

Incredibly complex ballistic missile defense computational problems have been a rich source of technological advance. The BMD challenge will stimulate supercomputer development through the 1980's.     

Errors in vector processing and the library libavi.a

In this paper, we describe the results of several tests that check the accuracy of numerical computation on the Cray supercomputer in vector and scalar modes. The known tests were modified to identify the critical point where roundings start causing problems. After describing the tests, we present an interval library called libavi.a. It was developed in Fortran 90 on the Cray Y-MP2E supercomputer of UFRGS-Brazil. This library makes interval mathematics accessible to the Cray supercomputers users. It works with real and complex intervals and intervals matrices and vectors. The library allows overloading of operators and functions. It is organized in four modules: real intervals, interval vectors and matrices, complex intervals, and linear algebra applications.     

一类国产复杂异构系统上的HPCG并行算法及高效实现研究

HPCG基准测试程序是一种新的超级计算机排名度量标准.该测试基准主要用于衡量超级计算机解决大规模稀疏线性系统的能力,更贴近实际应用,近年来广受关注.基于国产超级计算机研究异构众核并行HPCG软件具有非常重要的意义,其不仅可以提升国产超级计算机HPCG的排名,还对很多应用提供了并行算法、优化技术等方面的参考.本文面向某国产复杂异构超级计算机开展研究,首先采用了分块图着色算法对HPCG进行并行,并提出一种适用于结构化网格的图着色算法,该算法并行性能高于传统的JPL、CC等算法,且着色质量高,运用于HPCG后,迭代次数减少了3次,整体性能提升了6%.本文还分析了复杂异构系统各个部件传输的开销,提出一套更适用于HPCG的任务划分方法,并从稀疏矩阵存储格式、稀疏矩阵重排、访存等角度开展了细粒度的优化.另外在多进程计算时,还采用了内外区划分算法将核心函数SpMV、SymGS中的邻居通信操作进行了隐藏.最终整机测试时,性能达到国产超级计算机峰值性能的1.67%,相比单节点,整机弱可扩展性并行效率达到了92%.     

国产异构系统上的HPCG并行算法及高效实现

HPCG基准测试程序是一种新的超级计算机排名度量标准.该测试基准主要用于衡量超级计算机解决大规模稀疏线性系统的能力,更贴近实际应用,近年来广受关注.基于国产超级计算机研究异构众核并行HPCG软件具有非常重要的意义,其不仅可以提升国产超级计算机HPCG的排名,还对很多应用提供了并行算法、优化技术等方面的参考.面向某国产复杂异构超级计算机开展研究,首先采用了分块图着色算法对HPCG进行并行,并提出一种适用于结构化网格的图着色算法.该算法并行性能高于传统的JPL、CC等算法,且着色质量高,运用于HPCG后,迭代次数减少了3次,整体性能提升了6%.分析了复杂异构系统各个部件传输的开销,提出一套更适用于HPCG的任务划分方法,并从稀疏矩阵存储格式、稀疏矩阵重排、访存等角度开展了细粒度的优化.在多进程计算时,还采用内外区划分算法将核心函数SpMV、SymGS中的邻居通信操作进行了隐藏.最终整机测试时,性能达到了国产超级计算机峰值性能的1.67%,与单节点相比,整机弱可扩展性并行效率达到了92%.     

Second Generation of Vector Supercomputers

Integral scalar processing and increased parallelism are major advances in supercomputer evolution. Three second-generation machines—BSP, Cyber 205, and Cray-1— are here described and compared.     

Hybrid hierarchy storage system in MilkyWay-2 supercomputer

With the rapid improvement of computation capability in high performance supercomputer system, the imbalance of performance between computation subsystem and storage subsystem has become more and more serious, especially when various big data are produced ranging from tens of gigabytes up to terabytes. To reduce this gap, large-scale storage systems need to be designed and implemented with high performance and scalability.MilkyWay-2 (TH-2) supercomputer system with peak performance 54.9 Pflops, definitely has this kind of requirement for storage system. This paper mainly introduces the storage system in MilkyWay-2 supercomputer, including the hardware architecture and the parallel file system. The storage system in MilkyWay-2 supercomputer exploits a novel hybrid hierarchy storage architecture to enable high scalability of I/O clients, I/O bandwidth and storage capacity. To fit this architecture, a user level virtualized file system, named H²FS, is designed and implemented which can cooperate local storage and shared storage together into a dynamic single namespace to optimize I/O performance in IO-intensive applications. The evaluation results show that the storage system in MilkyWay-2 supercomputer can satisfy the critical requirements in large scale supercomputer, such as performance and scalability.     

A parallel, scalable, microprocessor-based database computer forperformance gains and capacity growth

A multiback-end database supercomputer, or MDBS, that embodies the architecture and performance of an experimental database computer and a number of database processors and their corresponding database stores, is discussed. Two studies are described. One focused on the design and architecture of the microprocessor-based MDBS, and the other on the performance and benchmark results for various loads and configurations     

现代超级计算机系统

本文论述了现代超级计算机系统发展的基本情况,深入分析了现代超级计算机系统的分类、特点与发展趋势。最后展望了90年代的发展前景。     

超级计算机的多层架构抽象及描述

超级计算机系统发展迅猛,不仅速度上日新月异,而且其结构也日益复杂化、多样化。为了能够准确有效地描述超级计算机的结构,给出统一的描述方法,提出超级计算机的多层架构思想。为超级计算机的进一步逻辑抽象或编写其模拟器提供了基础,并对超级计算机系统的设计具有一定的指导作用。     

Visualizing chaos: Lyapunov surfaces and volumes

The use of a graphic technique to visualize chaos for a dual-parameter one-dimensional map is described. It involves plotting the Lyapunov exponent with both height and color in a three-dimensional map, as a function of the two parameters. Color is determined using a geographic lookup table. A graphics supercomputer can rotate the map in real time. The technique demonstrates graphically interesting behavior in chaotic systems     

Load balancing methods and parallel dynamic programming algorithm using dominance technique applied to the 0–1 knapsack problem

The parallelization on a supercomputer of a one list dynamic programming algorithm using dominance technique and processor cooperation for the 0–1 knapsack problem is presented. Such a technique generates irregular data structure, moreover the number of undominated states is unforeseeable. Original and efficient load balancing strategies are proposed. Finally, computational results obtained with an Origin 3800 supercomputer are displayed and analyzed. To the best of our knowledge, this is the first time for which computational experiments on a supercomputer are presented for a parallel dynamic programming algorithm using dominance technique.     

基于GOS的远程实时可视化系统设计与实现*

提出了一个并行计算程序的远程实时可视化系统.通过实时可视化处理计算程序在计算过程中输出的数据文件,实现了对计算程序计算过程的可视化跟踪和实时分析;同时该系统采用GOS作为网格中间件,屏蔽了复杂的异构环境.     

一种求解汽车外流场问题的可扩展数值算法

受外型复杂、雷诺数高等因素影响,汽车外流场流动的数值计算规模巨大且难以精确求解。发展高效并行算法以利用超级计算平台资源来数值求解外流问题成为该领域的研究热点。文章提出一种全隐格式的可扩展并行Newton-Krylov-Schwarz算法对某真实汽车的外流场流动问题进行计算。通过与风洞试验以及主流计算流体力学软件的计算结果对比验证了算法的正确性。并行数值计算结果显示,文章的算法在数千处理器规模下仍具有很好的并行可扩展性。     

Scientific visualization in computational aerodynamics at NASA AmesResearch Center

Tools specifically oriented toward flow analysis problems are described. The approach is oriented toward performing graphics calculations on the supercomputer and using high-end workstations as rendering engines to display flow fields generated by aircraft flight. A suite of software provides both display and interactive analysis capability. Also described is animation hardware featuring Winchester disk technology, which allows the storage of more than one gigabyte of data and supports editing, special effects, and video output     

Distributed visualization at Los Alamos National Laboratory

Several approaches to visualizing simulated data are described. It is noted that a centralized video animation facility represents the state of the art in scientific visualization, but is too expensive to replicate widely. Methods used at LANL for connecting supercomputers with display units, ranging from dumb terminals to powerful workstations, are also described. Of particular interest is the Scientific Visualization Workbench, an inexpensive and effective way of using `televisualization' to move data from supercomputer to user and to modify the user's display. The implication of higher speed networks are discussed     

The μVP 64-bit vector coprocessor: a new implementation ofhigh-performance numerical computation

The architecture and design of the μVP, a single-chip vector coprocessor developed to meet the needs of high-performance processors, are described. The μVP is a supercomputer component implemented on a single large-scale-integrated (LSI) CMOS chip. With 206 MFLOPS single-precision and 106-MFLOPS double-precision performance at 50 MHz, the μVP offers a rate almost equivalent to that typical minisupercomputers