Parallel components for PDEs and optimization: some issues and experiences

Abstract

High-performance simulations in computational science often involve the combined software contributions of multidisciplinary teams of scientists, engineers, mathematicians, and computer scientists. One goal of component-based software engineering in large-scale scientific simulations is to help manage such complexity by enabling better interoperability among codes developed by different groups. This paper discusses recent work on building component interfaces and implementations in parallel numerical toolkits for mesh manipulations, discretization, linear algebra, and optimization. We consider several motivating applications involving partial differential equations and unconstrained minimization to demonstrate this approach and evaluate performance.

A Component Architecture for High-Performance Scientific Computing

“…Experiments have shown that the overheads for calls between directly connected CCA components are small and quite manageable in the context of scientific computing Norris et al 2002;. Calls using simple data types (those requiring no translation by Babel) cost the equivalent of a C++ virtual function call, exactly what is expected for the direct connection implementation described above.…”

“…• Norris and coworkers (Norris et al 2002) have compared a CCA component implementation and a non-component library implementation of a nonlinear unconstrained minimization problem solved using an inexact Newton method with a line search (Norris et al 2002). Each iteration of the Newton method required function, gradient, and Hessian evaluations, as well as an approximate solution of a linear system of equations.…”

Performance and Scalability of a Component-Based Grid Application

“…XCAT3 [12] is another CCA implementation that supports components distributed over different address spaces and accessible as collections of Grid services compliant to OGSI (Open Grid Services Infrastructure). In [16], CCA/Ccaffeine is used to componentise simulation software for partial differential equations. Components are produced by creating thin wrappers over existing numerical libraries.…”

Coupled Fusion Simulation Using the Common Component Architecture

“…The amount of extra code needed to acclimate a software module to the CCA environment can be as little as one extra (typically short) method, and several additional calls that manage ports provisioning and use. Experience has shown that componentization of existing software in the CCA environment is straightforward when starting from well-organized code [10].…”