Priorities and strategies, Los Alamos computer science institute.
R. R. Oldehoeft · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 2004
On March 18-19, 2002 the Los Alamos Computer Science Institute (LACSI) Executive Committee and Principal Investigators met to discuss methods of addressing issues raised in the 2001 LACSI Contract Review. The body was tasked to develop priorities and strategies to meet future programmatic and LANL computer science needs. A framework was developed to address long-term strategic thrust areas. Specific objectives were called out as near-term priorities. The objectives were folded into the framework to form a coherent planning view. On both April 8-9, 2003 and February 19-20, 2004, the LACSI Executive Committee and Principal Investigators met with senior LANL personnel to revise the framework, priorities, and strategies established at the planning meeting in 2002. The current framework outlines five strategic thrust areas: Components, Systems, Computational Science, Application and System Performance, and Computer Science Community Interaction. This document presents the research vision and implementation strategy in each of these areas. The goal of the component architectures effort is to make application development easier through the use of modular codes that integrate powerful components at a high level of abstraction. Through modularization and the existence of well-defined component boundaries (specified by programming interfaces), components allow scientists and software developers to focus on a their own areas of expertise. For example, components and modern scripting languages enable physicists to program at a high level of abstraction (by composing off-the-shelf components into an application), leaving the development of components to expert programmers. In addition, because components foster a higher level of code reuse, components provide an increased economy of scale, making it possible for resources to be shifted to areas such as performance, testing, and platform dependencies, thus improving software quality, portability, and application performance. A fundamental problem with this vision is that Los Alamos application developers, and most others in science, cannot afford to sacrifice significant amounts of performance for this clearly useful functionality. Therefore, an important part of the effort is to explore integration strategies that perform context-dependent optimizations automatically as a part of the integration process.