Support Request for Computational Science and Engineering Transitional Activities, 15-9069

Printer Friendly Version

Principal Investigators
Ernest A. Franke
Christopher J. Freitas
Richard L. Murphy
J. David Winningham

Inclusive Dates: 01/08/98 - 07/08/00

Background - Distributed parallel computing has become the dominant approach to solving complex scientific and engineering problems. The introduction of advanced CPUs such as the AMD Athlon™ will further improve the price and performance ratio, which in turn makes a network of workstations (NOWs) a truly affordable supercomputer. Clearly, advanced computational and data visualization techniques are essential to research at the Institute. Fortunately, distributed parallel processing provides the key to both affordable computing and effective visualization. This program is intended to be the first step toward providing Institute-wide resources and a cadre of trained staff members ready to create more cost-effective modeling, data mining, and simulation solutions for our clients.

Approach - SwRI senior management recognized the importance of computation science and engineering as a discipline fundamental to many Institute activities and funded the creation of a four-node, high-speed network in 1996. This network was based on first-generation Asynchronous Transfer Mode (ATM) technology and provided valuable experience with a heterogeneous set of workstations. Subsequently, funding was provided to greatly expand the ATM network and the number of workstations (37) supported. This program was instrumental in the migration of the SwRI backbone network to an ATM-based architecture. The three general targets of this program can be summarized as facilitation, education, and documentation. The program activities are classified as follows:

  • Facilitation
    • Establishment of a uniform Parallel Virtual Machine (PVM) base across all clustered workstations
    • Compilation of a comprehensive library of CSE software tools
    • Construction of a CSE 'intranet' Web server to provide the most current documentation to Institute staff
    • Importation or development of computing resource management tools
    • Production of a marketing brochure for use by any division.
    • Support design of an advanced, high-performance computer network within the Institute
    • Seek external funding for high-performance networking and CSE research programs
        
  • Education
    • Inititate a training program for Institute staff
        

  • Documentation
    • Document ATM network and workstation configuration

    • Examine technological trends shaping our CSE infrastructure and write a white paper projecting future directions

    • Create and document a method for integrating CSE support into projects

    • Specification of standards for future expansion of distributed parallel computing and advanced visualization for project support and competitiveness

Accomplishments - This program operated for ten quarters, including two no-cost extensions. A synopsis of activities touches upon all the targeted areas and includes other project activities derived from this program. SwRI's high-performance network infrastructure continues to expand, partially due to the stimulus of this program. The need to add or maintain computers connected to the DCF has lead to additional ATM switches being deployed in the Mechanical and Materials Engineering Division. More than 30 workstations can now participate as computing or visualization nodes within the DCF, although not all are available at one time.

An extended course called Introduction to Distributed Parallel Computing was offered through the Staff Development Office in the 4QCY1999. This course consisted of five weeks of lectures on PVM, network technologies, and techniques for writing parallel applications. From June 1 – 24 in CY2000, a course was presented on advanced visualization techniques, which are generally needed to understand properly the results produced by many applications.

The end result of this training is that twelve SwRI staff members attempted to use parallel-computing techniques within projects. The most successful of these attempts involved an internal research project for space weather modeling and a joint NSF/CONACyT-funded program involving Mechanical and Materials Engineering Division and the Monterrey Institute of Technology.

This program has fostered awareness of the power of distributed parallel computing and visualization techniques at a time when these areas are growing in importance. The Space Science and Engineering Division and Mechanical and Materials Engineering Division plan to build a homogeneous clustered computer on the Beowulf model in order to give PVM users an easier environment in which to work. The Institute has joined the Internet 2 consortium, which has a major goal of creating grid-computing technologies to attack very large scientific problems. Grid-computing refers to distributed parallelism for computing or information storage over a wide area rather than at a single laboratory. The Institute must now move to the next level of parallel computing education and must actively pursue collaborative research efforts in this area.

Intelligent Systems, Advanced Computer and
Electronic Technology, and Automation Program
2000 IR&D Home SwRI Home