[Next] [Previous] [Up] [Top] [Contents]

PDELab: A Development and Problem Solving Environment for
PDE-Based Applications on Parallel Machines

Principal Investigators: Elias N. Houstis, John R. Rice

Research Associates: A. Hadjidimos, A. Catlin, S. Weerawarana, M. Gaitatzes, S. Zhang, V. Vemkos, S. Markus, P. McCombs, K. Pantazopoulos, N. Houstis, X. Chen, N. Ramakrishna

Sponsors: AFOSR, ESPRIT, Intel, NASA, NSF, PRF, ARPA

Background

The trends in computer technology for large scientific computation are characterized by the diffusion of multiple processors cooperating synchronously or asynchronously in the execution of an application program. Unlike for sequential vector processors, algorithm implementation on complex parallel architectures is not logically consistent with traditional programming approaches and the scientific/engineering community is faced with the problem of exploring prospective solution strategies to achieve the high potential performance of parallel machines. Unfortunately this new hardware technology, despite its great potential, has widened the gap between hardware and applications. This project is a successor to Parallel ELLPACK.

One of the objectives of this project is to study the above problem and develop a Problem Solving Environment (PSE) that tries to close this gap, at least for certain applications that are governed by Partial Differential Equations (PDEs). Specifically, one of the problems we address involves the following fundamental research question:

One objective of this project is to develop a Problem Solving Environment (PSE) that tries to close the gap between hardware and applications that are governed by Partial Differential Equations (PDEs). We address the following fundamental research question:

How does one create a software system where PDE problems and parallel algorithms can be designed and specified in a machine-independent form within a reasonable time and that provides good implementation mappings to various parallel computer organizations?

Overall Design Objectives

One of the main objectives of this system is the creation of a machine-independent programming environment in which PDE problems and parallel PDE solvers can be easily implemented and executed. In order to realize such a computational environment, we need to establish fixed data structures and interfaces among the various parts of the PDE computations suitable for the implementation of parallel algorithms on various parallel machines.

Parallel processing of an application requires the partitioning and allocation of the underlying computation for the targeted architecture. This is usually an intractable problem. It is unrealistic to expect its solution by the novice user. This system provides automatic techniques for handling the mapping problem transparently and an interactive facility that allows the users to visualize such mappings, modify them, or specify new ones.

Very high-level application-oriented languages are becoming popular human-machine interfaces. PDELab provides a PDE specification language for describing the problem and its solution requirements textually. This interface is combined with a geometry specification tool for defining the PDE domain and various operations to be performed on it.

We expect the system to be used for PDE modeling in a production, educational, and research environment. Thus, it is necessary for the user to have control over various parameters of the solution process. In this system, the user is able to control the mesh, geometry partition, and mapping of the computation to the supported targeting architecture.

Background
Overall Design Objectives
CS Annual Report - 19 APR 1996

[Next] [Previous] [Up] [Top] [Contents]