ITR/SY: Data Parallel Component Software Components

University Of Utah, Salt Lake City UT

Investigators

Abstract

This project envisions an environment where a scientist can easily combine existing software components to create an efficient, highly parallel scientific simulation. To accomplish this, it will develop new software component technologies for high performance computing environments, including a high performance component framework and a new parallel Interface Definition Language (IDL). It will use those tools to investigate mechanisms for integrating high performance (often data parallel) computing with distributed software components. Success of this project will be measured by providing acceptable performance on at least two scientific applications using the new component framework. The applications chosen for the first test of this work (whose development has already started under separate funding) are bioelectric source localization and a combustion simulation of a high-energy material placed in a pool fire. Both applications will stress the capabilities and efficiencies of the proposed work. As software systems, including high performance parallel programs, become more complex, the use of component architectures can help manage many aspects of this complexity. However, such mechanisms are not prevalent in parallel scientific computing applications due to limitations of existing component systems. These include inefficient handling of large amounts of data, inability to express parallelism in the data or within a component, and difficulty matching parallelism in one component with parallelism in another. The new component architecture research will help overcome many of these limitations. It will, among other features, provide efficient mapping of components to processors, provide efficient communication between parallel components, facilitate migration of components (including checkpoint/restart capabilities), and facilitate redistribution of components for dynamic load balancing.