Adalsteinsson, Helgi - The Particle Simulation Toolkit. A toolkit for rapid development of massively parallel particle method simulations

ECCB 2002 Poster sorted by: Author | Number

Next | Previous poster (in order of the view you have selected)

Title: The Particle Simulation Toolkit. A toolkit for rapid development of massively parallel particle method simulations	P1
Adalsteinsson, Helgi hadalst@ca.sandia.gov Sandia National Laboratories, 7011 East Ave., Livermore, CA 94550, USA

Background: Computational chemistry and biochemistry have, over the past several years, become increasingly dependent on parallel computing. Since inclusion of parallelism adds substantial complexity to application development, many scientists opt to develop their programs using traditional sequential programming and retrofit the completed program with some parallel capabilities. The end result of such parallelization retrofits is all to often application programs that do not see significant speed gains past a few nodes in a distributed computation environment.
One of the more exciting additions to application development in computational chemistry and biochemistry is the introduction of modeling toolkits that allow developers to assemble computational components in an interactive manner. These toolkits have been designed to give users access to force field simulations (e.g. the Molecular Modeling Toolkit MMTK), to interface existing simulation and visualization packages (e.g. CAMPOS), and to access a variety of file formats and online databases used in molecular biology (e.g. Biopython). The most ambitious computational chemistry development environment to date is the Molecular Operating Environment (MOE), which provides extensive functionality for computational methods development. MOE can, however, not be classified as a component toolkit, since application developers can not use components from MOE in standalone applications.
Although the toolkits mentioned above provide an excellent route for rapid application development and testing, their computational performance is generally insufficient for the development of applications with serious computational requirements. Only MOE provides any parallel functionality, which is restricted to distributing independent tasks (invoked through remote execution, no message passing or synchronized parallelism). To create massively parallel applications, the developer is, therefore, forced to re-implement the program before distributing it to end-users.

Goal: The Particle Simulation Toolkit (PST) is designed to be an interactive application development environment that provides highly optimized parallel components for application development in particle simulations, with particular emphasis on computational chemistry and biochemistry. In order to achieve these goals, PST defines a framework for parallelization and the addition of specialized application components, as well as a collection of standard components.
PST supports a mixture of shared memory and distributed (message passing) parallelism, which enables it to take full advantage of clusters of multi-processor SMP machines, such as the newer and emerging ASCI computing platforms. The framework also provides components for handling mixtures of independent tasks and parallel applications that run on a subset of the available nodes, as well as a system for wrapping and calling legacy applications. Current efforts are focused on integrating the Massively Parallel Quantum Chemistry program MPQC and high performance parallel optimization packages such as TAO, Opt++, and Appspack, and adding capabilities for solid state simulations and smart particle methods.