一种新的MPI Allgather算法及其在万亿次机群系统上的实现与性能分析

给出一个新的MPI Allgather算法--邻居交换算法（neighbor exchange）.提出的平均逻辑通信距离的概念和计算公式,可以有效地衡量通信的局部性.通过分析,发现在4种MPI Allgather算法中,邻居交换和环算法均具有最优的通信局部性.在万亿次机群深腾6800和曙光4000A上对4个MPI Allgather算法进行的性能测试和分析结果表明,邻居交换算法的长消息通信性能最优,中长消息通信性能不稳定,短消息通信性能次于递归倍增和Bruck算法.     

航班进港过程建模与控制

控制和监视航班间隔是空中交通管制系统的主要职责,新一代航空运输系统要逐步实现空中交通管制的自动化.探讨建模和最小方差控制在实现进港航班时间间隔控制中的应用.在研究空中交通管制系统运行机制的基础上,确定航班进港过程的基本属性为离散事件动态系统.讨论了离散事件动态系统分析设计各种方法用于进港航班调度的利弊.对航班进港过程实测数据进行分析,确定该过程具有低通特性.利用系统辨识方法建立了航班进港过程的受控模型.采用最小方差控制方法实现了对过程的控制,证明了通过辨识建立的航班进港过程模型是合理的,为航班进港过程管制自动化的工程实现给出了新的思路和方法.     

Alltoall通信性能模型研究

Alltoall是一种重要的MPI(message passing interface)集合通信类别,是影响许多并行程序并行效率的重要因素。但对于大规模并行计算机上Alltoall集合通信的评测和理论分析仍较为缺乏,导致许多应用程序的通信模块设计和使用不合理。首先,开展了MPI基本通信性能的测试和分析,发现随着MPI进程数的增加,其性能波动也增加,而这种波动源自网络竞争。为此,在传统的Alltoall性能评估模型中引入了网络竞争因素,新模型不仅考虑传统的通信带宽和通信延迟参数,还考虑了通信竞争因素。某国产并行机平台上的测试结果显示:引入网络竞争模型的新Alltoall性能评估模型可以较为准确地预估Alltoall性能,体现出网络竞争开销对Alltoall性能的影响。     

基于InfiniBand的RPC优化

随着无限带宽(IB)网络技术的发展,IB网络以其高性能、低延迟以及低功耗和低成本的优势逐渐在数据中心占据重要的地位.而作为重要的网络通信模型,远程过程调用(RPC)协议如何利用IB网络的远程内存直接访问RDMA特性是一个重要议题.NFSover RDMA解决了内核态下网络文件系统(NFS)利用RDMA协议传输数据的问题.然而,面向跨越内核态和用户态的RPC应用,目前仍然没有基于IB的RPC相关优化.通过将IB协议族中的套接字直接协议(SDP)集成到现有RPC协议中,使内核态以及用户态RPC程序都能使用IB网络的RDMA协议,从而提升网络传输性能.经测试,此方案与传统的基于TCP/IP协议的RPC实现相比,用户态RPC与内核态RPC可分别获得10％00和35％的性能提升.     

国产百万亿次机群系统Alltoall性能测试与分析

随着高性能计算机的应用和发展,并行应用程序所使用的处理器数越来越多,进程间的通信量也不断增多,这对应用程序的性能有很大影响. 在采用一种快速傅里叶变换HFFT对曙光5000A进行性能测试时发现,MPI集合通信函数MPI Alltoall的巨大通信开销是并行程序设计的瓶颈.为此,对现有主流Alltoall算法在曙光5000A和深腾7000上进行性能测试与分析,以期对未来的Alltoall算法的优化工作做出贡献.利用不同消息长度和不同进程数测试了Alltoall函数多种算法的性能,这些算法包括二维网格算法、三维网格算法、Bruck算法、原始算法、成对交换算法、递归倍增算法、环算法以及LAM/MPI中的简单算法等.实验结果表明:消息长度较小时,在曙光5000A上采用原始算法和Bruck算法的性能较好,而在深腾7000上用时较少的算法是简单算法和Bruck算法;对于长消息,曙光5000A上最优的算法是环算法,深腾7000上成对交换性能最优.     

基于统计信息的聚类边界模式检测算法

为有效地检测聚类的边界点,提出基于统计信息的边界模式检测算法。根据数据对象的k距离统计信息设定邻域半径,再利用对象邻域范围内邻居的k距离统计信息寻找边界点。实验结果表明,该算法可以有效地检测出任意形状、不同大小和不同密度聚类的边界点,并可以消除噪声。     

Fast routing computation on InfiniBand networks

The InfiniBand architecture has been proposed as a technology both for communication between processing nodes and I/O devices, and for interprocessor communication. Its specification defines a basic management infrastructure that is responsible for subnet configuration and fault tolerance. Each time a topology change is detected, new forwarding tables have to be computed and uploaded to devices. The time required to compute these tables is a critical issue, due to application traffic is negatively affected by the temporary lack of connectivity. In this paper, we show the way to integrate several routing algorithms, in order to combine their advantages. In particular, we merge a new proposal, characterized by its high computation speed but low efficiency, with a traditional one, slower but more efficient. Our goal is to provide new routes in a short period of time, minimizing the degradation mentioned before, and maintaining, at the same time, high network performance.     

IBA与企业存储网络

介绍了网络存储的两种机制,展望了在一种新的I／O体系结构InfiniBand上实现网络存储的前景。     

InfiniBand技术分析与应用研究

主要对InfiniBand技术的结构、原理进行了较为详细的分析和研究,同时指出了该技术的优点和存在的问题。     

基于InfiniBand的高性能计算机技术研究

网络性能一直是制约高性能计算技术发展的瓶颈,无论是面向计算的网络还是面向存储的网络,通信速度的发展远落后于CPU的发展．InfiniBand互连结构能够缩短网络和CPU之间的性能差距,使高性能计算机的性能趋于平衡．2000年在InfiniBand协议发展初期,国家并行计算机工程技术研究中心就在国家“八六三”计划的支持下开始对InfiniBand协议展开了深入研究,旨在改进高性能计算特别是集群系统的互连性能,研制出符合InfiniBand标准规范的高性能互连部件,并最终开发出了具有自主知识产权的InfiniBand互连网络产品．论述了以自行研制的InfiniBand部件技术为基础的高性能集群计算机系统的组成、结构和应用,并对系统性能进行了实验分析．     

基于InfiniBand的存储系统目标模拟器

本文设计了一个支持SRP数据旁路加速和多I／O并发处理的存储系统目标模拟器I／O存取模型,通过主机通道GUID标识号与逻辑卷之间的动态映射关系实现存储系统逻辑存储资源的授权访问与屏蔽策略,并进一步提出了通过并发访问阵列设备来增加I／O带宽的方案。测试证明,目标器具有高带宽、低延时、可扩展、卷映射灵活等特点。     

InfiniBand网络胖树路由算法分析

胖树拓扑结构是搭建集群和HPC比较流行的结构,它具有无阻塞传输和对分带宽的优点。本文介绍了In-finiBand子网管理和胖树结构,分析了胖树路由算法的计算过程和相关改进。     

基于过程模式的软件过程框架

Process pattern is an excellent method to express software process knowledge, it can express process knowledge in different granularity and realize the non-gap connection of process knowledge, it supports the reuse and persistent improvement of process knowledge. In this paper, we put forward a process Pattern Based Software Pro-cess Framework(PB-SPF), its three layer architecture has realized the high abstraction of software process and theseparation of its contents, so it have good reusability and adaptability. We can use it as a base for the research of soft-ware process, the building and enactment of process model.     

Multi-server queueing system with a Batch Markovian Arrival Process and negative customers

A multi-server queueing system with a finite buffer, where a Batch Markovian Arrival Process (BMAP) arrives, is studied. The servicing time of a customer has a phase-type (PH) distribution. Customers are admitted to the system in accordance with the disciplines of partial admission, complete admission and complete rejection. Except standard (positive) customers, a MAP flow of negative customers arrives to the system. In a random way, a negative customer removes from the system one of the positive customers that are at the server. A stationary distribution of system state probabilities, the Laplace-Stieltjes transformation of the stationary distribution of waiting time, major characteristics of system performance are found. Numerical examples are given.     

A Multiserver Retrial Queueing System with Batch Markov Arrival Process

A sufficient condition for a BMAP/M/N system to have a stationary state probability distribution and an algorithm for computing this distribution are investigated.     

The distributed virtual shared-memory system based on the InfiniBand architecture

Even though there have been strong research activities about distributed virtual shared-memory (DVSM) systems, their architectures have been not widely used in current high-performance computing markets. The reason is that the previously introduced DVSM systems use conventional interconnection technologies like Ethernet, which incurs high execution overhead due to process interruption at data communication for memory consistency. In this paper, we present the DVSM architecture based on the next generation of an interconnection technique, the InfiniBand Architecture (IBA). Because the IBA supports shared-memory programming semantics by means of remote direct-memory access (RDMA) and atomic operations in hardware, we can minimize the communication overhead for memory consistency on the DVSM system. For characterizing multithreaded applications on our IBA-based DVSM system, we examined two different shared-memory programming models, i.e. SPMD and OpenMP benchmarks. We show that our DVSM to use full features of the IBA can improve the performance significantly over the IPoIB-based DVSM system in all benchmarks, and also comparable to the bus-based shared-memory multiprocessor system in some benchmarks.     

基于IBA构建高性能的MPI环境

IBA作为一种I/O间、主机间的下一代高速互联方式,在高性能计算领域越来越成为瞩目的焦点。消息传递接口MPI已经成为并行程序设计事实上的国际标准。该文详细介绍了在IBA之上构建一个高性能的MPI环境的方法,并对消息延迟和消息带宽进行了测试,对测试数据进行了分析。     

A methodology to enable QoS provision on InfiniBand hardware

The Journal of Supercomputing - Modern high-speed interconnection networks include support for the provision of quality of service (QoS) to the applications. The output scheduling algorithm plays...     

基于模式树的XETL过程研究

XML数据与传统的关系型数据存在的差异,使得传统数据仓库的ETL方法已经不适用于XML数据,而目前也没有专门的、有效的适用于XML数据的ETL方法。针对这一问题,提出基于模式树的XML转换处理过程——XETL。从数据模型和谓词模式研究XETL模型,基于XETL模型定义ETL过程中属性选择、空置处理、聚合以及属性重命名4类主要的转换处理操作。