Catálogo de publicaciones - libros

The university education in parallel and high-performance computing often suffers from a significant gap between the effects and potential performance taught in the lectures on the one hand and those practically experienced in exercises or lab courses on the other hand. With a small number of processors, the results obtained are often hardly convincing; however, supercomputers are rarely accessible to students doing their first steps in parallel programming. In this contribution, we present our experiences of how a state-of-the-art mid-size Linux cluster, bought and operated on a department level primarily for education and algorithm development purposes, can be used for teaching a large variety of HPC aspects. Special focus is put on the effects of such an approach on the intensity and sustainability of learning.

In this paper we present our approach to teaching High Performance Computing at both the undergraduate and graduate level. For undergraduate students, we emphasize the key role of an hands on approach. Parallel computing theory at this stage is kept at minimal level since this knowledge is fundamental, but our main goal for undergraduate students is the required ability to develop real parallel applications. For this reason we spend about one third of the class lectures on the theory and remaining two thirds on programming environments, tools and libraries for development of parallel applications. The availability of widely adopted standards provides us, as teachers of high performance computing, with the opportunity to present parallel algorithms uniformly, to teach how portable parallel software must be developed, how to use parallel libraries etc. When teaching at the graduate level instead, we spend more time on theory, highlighting all of the relevant aspects of parallel computation, models, parallel complexity classes, architectures, message passing and shared memory paradigms etc. In particular, we stress the key points of design and analysis of parallel applications. As a case study, we present to our students the parallelization of a real computational science application, namely a remote sensing SAR (Synthetic Aperture Radar) processor, using both MPI and OpenMP.

The cross fertilization of methods and techniques between different subject areas in the undergraduate curriculum is a challenge, especially at the more advanced levels. This paper describes an attempt to achieve this through a tutorial based around a traditional high performance computing application, namely molecular dynamics. The tutorial exposes students to elements of software design patterns, the construction of graphical user interfaces, and concurrent programming concepts. The tutorial targets senior undergraduate or early postgraduate students and is relevant to both those majoring in computing as well as other science disciplines.

Multidisciplinary research reliant upon high-performance computing stretches the traditional educational framework into which it is often shoehorned. Multidisciplinary research centers, coupled with flexible and responsive educational plans, provide a means of training the next generation of multidisciplinary computational scientists and engineers. The purpose of this paper is to address some of the issues associated with providing appropriate education for those being trained by, and in the future being employed by, multidisciplinary computational science research environments.

Over three decades of parallel computing, new computational requirements and systems have steadily evolved, yet parallel software remains notably more difficult relative to its sequential counterpart, especially for fine-grained parallel applications. We discuss the role of education to address challenges posed by applications such as informatics, scientific modeling, enterprise processing, and numerical computation. We outline new curricula both in computational science and in computer science. There appear to be new directions in which graduate education in parallel computing could be directed toward fulfilling needs in science and industry.

Multiphysics simulations are increasingly playing a critical role in scientific and engineering applications. The complex and cross-disciplinary nature of such applications poses many challenges and opportunities in both research and education. In this paper we overview some of these research challenges, as well as an academic program designed to prepare students to meet them.

The design and deployment of Virginia Tech’s terascale computing cluster is described. The goal of this project is to demonstrate that world-class on-campus supercomputing is possible and affordable, and to explore the resulting benefits for an academic community consisting of both computational scientists and computer science researchers and students. Computer science research in high performance computing systems benefits significantly from hands-on access to this system and from close collaborations with the local computational science user community. We describe an example of this computer science research, in the area of dynamically resizable parallel applications.

The pricing of a portfolio of financial instruments is a common and important computational problem in financial engineering. In addition to pricing, a portfolio or risk manager may be interested in determining an effective hedging strategy, computing the value at risk, or valuing the portfolio under several different scenarios. Because of the size of many practical portfolios and the complexity of modern financial instruments the computing time to solve these problems can be several hours. We demonstrate a powerful and practical method for solving these problems on clusters using web services.

At UCLA’s Plasma Physics Group, to achieve accessible computational power for our research goals, we developed the tools to build numerically-intensive parallel computing clusters on the Macintosh platform. Our approach is designed to allow the user, without expertise in the operating system, to most efficiently develop and run parallel code, enabling the most effective advancement of scientific research. In this article we describe, in technical detail, the design decisions we made to accomplish these goals. We found it necessary for us to “reinvent” the cluster computer, creating a unique solution that maximizes accessibility for users. See: http://daugerresearch.com/.

The Green Grid is an ambitious project to create a shared high performance computing infrastructure for science and engineering at Dartmouth College. The Green Grid was created with the support of the Dean of the Faculty of Arts & Sciences to promote collaborative computing for the entire Dartmouth community. We will share our design for building campus grids and experiences in Grid-enabling applications from several academic departments.