Northeast Parallel Architecture Center

Benchmarking the Computation and Communication Performance of the CM-5

Kivanc Dincer et al. — Fri, 17 Dec 2010 06:39:13 PST

Thinking Machines' CM-5 machine is a distributed-memory, message-passing computer. In this paper we devise a performance benchmark for the base and vector units and the data communication networks of the CM-5 machine. We model the communication characteristics such as communication latency and bandwidths of point-to-point and global communication primitives. We show, on a simple Gaussian elimination code, that an accurate static performance estimation of parallel algorithms is possible by using those basic machine properties connected with computation, vectorization, communication, and synchronization. Furthermore, we describe the embedding of meshes or hypercubes on the CM-5 fat-tree topology and illustrate the performance results of their basic communication primitives.

Object Web (Java/CORBA) based RTI to support Metacomputing M&S

Geoffrey C. Fox et al. — Fri, 17 Dec 2010 06:39:11 PST

We present here our Pragmatic Object Web based approach to High Performance Modeling and Simulation and we describe the associated middleware software recently prototyped at NPAC: JWORB (Java Web Object Request Broker) which integrates HTTP and IIOP protocols, and Object Web RTI which implements DMSO RTI 1.3 on top of the JWORB based CORBA / Java software bus. We explain how JWORB and OW RTI are used to build WebHLA – an interactive FMS training environment and we outline our plan towards WebHLA based Virtual Prototyping Environments for Testing, Evaluation and Simulation Based Acquisition.

Parallel and Distributed Computing using Pervasive Web and Object Technologies

Geoffrey C. Fox et al. — Fri, 17 Dec 2010 06:39:08 PST

We review the growing power and capability of commodity computing and communication technologies largely driven by commercial distributed information systems. These systems are built from CORBA, Microsoft’s COM, Javabeans, and less sophisticated web and networked approaches. One can abstract these to a three-tier model with largely independent clients connected to a distributed network of servers. The latter host various services including object and relational databases and, of course, parallel and sequential computing. High performance can be obtained by combining concurrency at the middle-server tier with optimized parallel back-end services. The resultant system combines the needed performance for large-scale HPCC applications with the rich functionality of commodity systems. Further, the architecture with distinct interface, server and specialized service implementation layers, naturally allows advances in each area to be easily incorporated. We show that this approach can be applied to both metacomputing and to provide improved parallel programming environments. We describe exploitation issues within a CORBA context and illustrate how performance can be obtained within a commodity architecture. Examples are given from collaborative systems, support of multidisciplinary interactions, proposed visual HPCC ComponentWare, distributed simulation, and the use of Java in high-performance computing.

Design of an Application Development Toolkit for HPF/Fortran 90D

Manish Parashar et al. — Mon, 11 Oct 2010 10:28:09 PDT

The development of efficient application software capable of exploiting available High Performance Computing (HPC) systems is non-trivial and is largely governed by the availability of sufficiently high-level languages, tools, and application development environments. In this paper we describe the design and operation of a toolkit for HPF/Fortran 90D application development. The toolkit incorporates the following systems: (1) ESP: An Interpretive Framework for HPF/Fortran 90D Performance Prediction; (2) ESP-i: A HPF/Fortran 90D Functional Interpreter; and (3) ESPial: An Integrated Environment for HPF/Fortran 90D Application Development & Execution. The toolkit has been implemented on the iPSC/860 hypercube system, and is supported by an interactive, graphical user interface (ESPView) which provides application developers with the following functionality: design evaluation capability, functional verification capability, performance visualization support, experimentation capability,

Compiling distribution directives in a Fortran 90D compiler

Zeki Bozkus et al. — Mon, 11 Oct 2010 10:28:09 PDT

Data Partitioning and mapping is one of the most important steps of in writing a parallel program; especially data parallel one. Recently, Fortran D, and subsequently, High Performance Fortran (HPF) have been proposed to allow users to specify data distributions and alignments for arrays in programs. This paper presents the design of a Fortran 90D compiler that takes a Fortran 90D program as input and produces a node program + message passing calls for distributed memory machines. Specifically, we present the design of the Data Partitioning Module that processes the alignment and distribution directives and illustrate what are the important design considerations. We show that our compiler produces portable, yet an efficient code. We also present the performance of the code produced by the compiler and compare it with the performance of the hand written code. We believe, this design can be used by implementors of the HPF compilers.

Techniques for Empirical Testing of Parallel Random Number Generators

Paul D. Coddington et al. — Mon, 11 Oct 2010 10:28:08 PDT

Parallel computers are now commonly used for computational science and engineering, and many applications in these areas use random number generators. For some applications, such as large-scale Monte Carlo simulations, it is crucial that the random number generator have good randomness properties. Many programs are available for testing the quality of sequential random number generators, but very little work has been done on testing parallel random number generators. We present some techniques for empirical testing of random number generators on parallel computers, using tests based on computational science applications as examples. In particular, we focus on tests based on parallel algorithms developed for Monte Carlo simulations of the two dimensional Ising model, for which exact results are known. Preliminary results of these tests are presented for several parallel random number generators.

Collaborative Scientific Data Visualization

Byeongseob Ki et al. — Mon, 11 Oct 2010 10:28:08 PDT

We have designed a collaborative scientific visualization package that will aid researchers from distant, diverse locations to work together in developing scientific codes, providing them with a system to analyze their scientific data. We have utilized Java to develop this infrastructure. Two important areas which we have concentrated on developing are 1) a collaborative framework from which the scientific data is interpreted and utilized, and 2) a framework, which is customizable to the suit the needs of a particular task and/or scientific group.

Exploration of Emerging HPCN Technologies for Web-Based Distributed Computing

Hon W. Yau et al. — Mon, 11 Oct 2010 10:28:07 PDT

The surge in the popularity of the World Wide Web (WWW) has corresponded to a decreasing market for specialised high performance computers. This paper discusses how, by making use of technology developed from the broader end of the computing pyramid, much of the past decade's work in distributed computing can be realised in the context of the larger WWW market. Not only do these new technologies offer fresh possibilities, but their pace of development is unlikely to be matched by the traditional high performance research community. A motivating application, discussions of the pertinent emerging technologies, and NPAC's investigations of them, will be presented.

DARP: Java-based Data Analysis and Rapid Prototyping Environment for Distributed High Performance Computations

Erol Akarsu et al. — Thu, 07 Oct 2010 09:21:53 PDT

The integration of a compiled and interpreted HPF gives us an opportunity to design a powerful application development environment targeted for high performance parallel and distributed systems. This Web based system follows a three-tier model. The Java front-end holds proxy objects which can be manipulated with an interpreted Web client (a Java applet) interacting dynamically with compiled code through a tier-2 server. Although targeted for HPF back-end, the system’s architecture is independent of the back-end language, and can be extended to support other high performance languages.

Exploration of the InfoMall Concept "Building on the Electronic InfoMall"

Geoffrey C. Fox et al. — Wed, 06 Oct 2010 09:53:11 PDT

This document describes: the InfoMall concept; how it is employed by the Northeast Parallel Architectures center (NPAC) as a technology transfer program, how it could be used by Rome Laboratory and by the United States Air Force Materiel Command (US AFMC). A description of the “Electronic InfoMall” system built on the World Wide Web as a pilot project for Rome Laboratory is also given as well as some experiences building WWW systems for academic, commerce and industry. It should be emphasized that this document is primarily a description of the potential uses of the InfoMall concept and the human interactive processes involved in InfoMall and is not primarily about the HPCC technologies that make the InfoMall process work. These technologies are well described elsewhere although we summarize their main features in this document.

The Virtual Computing Environment

Philip Rousselle et al. — Wed, 06 Oct 2010 09:53:11 PDT

A network of supercomputers and high-performance workstations appears to be the only reasonable way to provide adequate computing resources for the Grand Challenge problems of the next century. Such a collection of computers and supporting software environments is called a virtual computing environment (VCE). This paper describes the motivation and goals of the VCE project, followed by a description of the system. The paper concentrates on the runtime aspects of the VCE, and concludes with a discussion of a small prototype system that has been built using the Isis distributed toolkit.

A Methodology For Developing High Performance Computing Models: Storm-Scale Weather Prediction

Nikos Chrisochoides et al. — Wed, 06 Oct 2010 09:53:10 PDT

A methodology for developing future generations of a storm-scale weather prediction model for Massively Parallel Processing is described. The forecast model is the Advanced Regional Prediction System (ARPS), a three-dimensional, fully compressible, non-hydrostatic predictive model. In the short term, the computational goals include developing a portable, scalable model for distributed memory SIMD and MIMD architectures, while preserving a high degree of modularity to support rapid design and validation, maintainability, educational goals and operational testing. Longer term computational goals include a parallel adaptive mesh refinement scheme. A FortranD/High Performance Fortran version of the ARPS provides portability in the current version of the model, and supports future model research goals.

Basic Issues and Current Status of Parallel Computing -- 1995

Geoffrey C. Fox — Tue, 05 Oct 2010 08:50:19 PDT

The best enterprises have both a compelling need pulling them forward and an innovative technological solution pushing them on. In high-performance computing, we have the need for increased computational power in many applications and the inevitable long-term solution is massive parallelism. In the short term, the relation between pull and push may seem unclear as novel algorithms and software are needed to support parallel computing. However, eventually parallelism will be present in all computers -- including those in your children's video game, your personal computer or workstation, and the central supercomputer.

Applications And Enabling Technology For Nynet Upstate Corridor

Salim Hariri et al. — Tue, 05 Oct 2010 08:50:19 PDT

Current advances in telecommunication and computing will have significant impact on the proliferation of high performance computing and communication (HPCC) applications. With these emerging technologies, it is feasible to run parallel and distributed applications across a high speed wide area network which was not possible a few years ago; the high latency and low bandwidth were the main bottlenecks for the wide area network-based computing. This has lead to the deployment of several high speed networks across the country (eg. NYNET). In this report, we describe some of the HPCC applications and our experiences and lessons learned from running them over the NYNET testbed. NYNET is one of the first wide area networks to use commercially available ATM switches and first to have an aggressive research plan to develop a wide range of large scale HPCC applications. NYNET testbed covers all the New York State and part of Massachusetts State and provides an interconnection between leading educational institutions, government laboratories and industrial labs. The main objectives of this project were to develop and demonstrate HPCC applications and evaluate current HPCC enabling technologies. We show the benefits that can be achieved from applying HPCC technologies to implement applications encountered in military (eg. Multi-target tracker), industry (eg. Financial modeling), scientific applications (eg. Electromagnetic scattering) and health care. Futhermore, we benchmark and evaluate several parallel and distributed platforms and software tools for developing such HPCC applications on NYNET.

Internetics: Technologies, Applications and Academic Field, or, Parallel Computing and Computational Science Do Not Quite Work

Geoffrey C. Fox — Tue, 05 Oct 2010 08:50:18 PDT

Ten years ago, we were all sure that parallel computing technology and the interdisciplinary academic field of computational science would be center pieces of both academic and economic growth. We show that this insight was, in principle, correct but was an incomplete vision for large-scale computation implies both increased computer power and increasing numbers of users and applications. Parallel computing undoubtedly works on essentially all problems, but we were unable to produce deployable software systems. Further, few industries could achieve adequate return to justify investment in parallel computers, except in a few areas such as databases. Computational science is the academic field on the interface of computer science with fields such as physics, chemistry, and applied mathematics. This expertise allows you to be very useful and, in principle, is an excellent area of study, but is not a wise field for many students as employers and universities prefer traditional fields. We show how parallel computing and computational science has evolved into Internetics, which is a vibrant growing and much larger field that surely does work both in principle and in practice. Internetics embodies the technologies and expertise used in building large-scale distributed systems and linking fields like physics not just with parallel computers, but with the Web of complex heterogeneous computers. This is CORBA and Java, and not just MPI and HPF. It is Internetics that is the emerging academic field, and not computational science, and internetics is of growing attraction to students and employers. Using an Internetics base, we will produce much better software environments for parallel systems, but the commercial and academic fields associated with parallelism will not grow in the near future. We argue that we almost "got it right" and the essential features of the original vision were correct and are part of current broader thrust.

From Computational Science to Internetics: Integration of Science with Computer Science

Geoffrey C. Fox — Tue, 05 Oct 2010 08:50:17 PDT

We describe how our world dominated by Science and Scientists has been changed revolutionized by technologies moving with Internet time. Computers have always been well-used tools but in the beginning only the science counted and little credit or significance was attached to any computing activities associated with scientific research. Some 20 years ago, this started to change and the area of computational science gathered support with the NSF Supercomputer centers playing a critical role. However this vision has stalled over the last 5 years with information technology increasing in importance. The Holy Grail of computational science-- scalable parallel computing-- is still important but is just one supporting component of the Internet revolution. We discuss the emergence of the field of Internetics-- bridging computer science and all application areas whether simulation or information based. Internetics is an exciting field, which seems complete and rich enough to be a lasting interdisciplinary area. Physics and other core science and engineering disciplines used to attract the very best minds but now their popularity is declining. We describe curricula initiatives that can reinvigorate these fields. This curricula turmoil must be addressed by our education infrastructure whose professional staff find it hard to develop courses to satisfy student and employer interests in times of such rapid change. Distance education is very relevant as it can be used to disseminate expertise to students and teachers in these new areas. All of this has implications for our educational institutions, which could be quite profound and will be

A Tale of Two Applications on the NII

Geoffrey C. Fox — Tue, 05 Oct 2010 08:50:17 PDT

We describe the expected capability of the NII (as an evolution of the Internet) interns of five broad service areas---collaboration, multimedia information dissemination, commerce, metacomputing, and Webtop productivity. We illustrate the demands on these services and the technology implications by examination of two application areas---manufacture of complex systems, such as aircraft and crisis management, command and control.

MPJ: A Proposed Java Message Passing API and Environment for High Performance Computing

Mark Baker et al. — Tue, 05 Oct 2010 08:50:16 PDT

In this paper we sketch out a proposed reference implementation for message passing in Java (MPJ), an MPI-like API from the Message-Passing Working Group of the Java Grande Forum [1,2]. The proposal relies heavily on RMI and Jini for finding computational resources, creating slave processes, and handling failures. User-level communication is implemented efficiently directly on top of Java sockets.

Common Runtime Support for High Performance Languages

Geoffrey C. Fox — Tue, 05 Oct 2010 08:50:15 PDT

Widespread adoption of parallel computing depends on the availability of improved software environments. An essential component of these environments will be high-level languages. Several languages for exploiting data-parallelism (or task-parallelism) have been developed, or are under development. The stated goal of this project has been to provide a public domain infrastructure for runtime support of these high-level languages. The targeted languages include parallel versions of Fortran and C++, but our intention has been to provide uniform runtime support for many source languages.

TANGO - a Collaborative Environment for the World-Wide Web

Lukasz Michal Beca et al. — Tue, 05 Oct 2010 08:50:15 PDT

Geographical and logical growth of the World-Wide Web is accompanied by a fast technological development. Web can be successfully used as a platform for implementation of diverse applications. Distributed and collaborative systems are among the most challenging Web applications. TANGO is an integration platform which enables implementation of Web-based collaborative environments. The system provides means for fast integration of Web- and non-Web-applications into one multi-user collaborative systems. In this paper we describe the functional model, requirements, system design and certain implementation issues of the TANGO system.