Research Assistants: T. Drashansky, S. Markus
The proper exploitation of massively parallel computers requires a software structure that ranges from the very high level, user friendly problem solving environments to the low level, very efficient computational kernels. The work here is to create part of this structure, a framework for solving partial differential equations (PDEs) which model the physics of flows, heat, magnetic fields, material properties, etc. Further, we will build an initial set of software kernels to flesh out this framework.
The context of the basic research is the development of a software environment and methodology for complex applications running on massively parallel platforms. We have selected a large scale application related to design and simulation of physical systems to demonstrate the scalability and efficiency of our approach: Electronic Prototyping for Physical Objects Design (EPPOD). We believe that EPPOD is one of the grand challenges for computer science which will require massively computational power and smart software for its realization. The design and implementation of smart and integrated numerical simulation systems on high performance computers requires the integration of many advanced information technologies from the fields of programming languages, databases and knowledge bases, user interface systems, software development tools, symbolic/numerical/AI techniques, and parallel and distributed processing. We aim to build a development environment for the construction of problem solving systems for large scale applications and massively parallel machines. It will provide an integrated and open ended toolkit to speed the development of large-scale simulation codes on parallel machines, promote reusability and facilitate easy maintenance of such systems. It will support the development of complex, intelligent applications as EPPOD through a machine-independent user interface based on module-oriented programming. Some of the components in this environment are software parts such as computational kernels or frameworks and modules of libraries implemented in various targeted architectures.
Creating a system for physical design and analysis, such as mechanical design, biomedical design, and so on, is one of the next "Grand Challenges" for computer applications. These systems will provide accurate computer simulations of physical objects coupled with powerful design optimization tools to allow prototyping and final design of a broad range of items. When such systems become a reality, they will have an even greater impact than systems for electronic design, one of the great achievements of computing technology of the past decade. Our studies indicate that the computer science problems arising from such systems can be surmounted. The deep challenges in systems integration, in obtaining enough computing power, in devising competent geometric tools, and so on can be met. It is clear that enough physical phenomena are well understood so that a useful, accurate, and broadly applicable system for physical design can be created. The principal computational problem underlying EPPOD is the solution of partial differential equations (PDEs) which model these physical phenomena. Thus, the challenge is to incorporate this knowledge into a high-level design system where the design parameters (e.g., shapes, materials, construction techniques, environment conditions) are optimized rapidly and accurately. It is reasonable to expect the appearance of powerful physical design and analysis systems within five or ten years.
CS Annual Report - 19 APR 1996