18. Parallel Methods and Software for Multicomponent Simulations

Abstract

Computational simulation is an increasingly important technique both for understanding natural phenomena and for designing engineered devices. Many engineering systems and physical phenomena of interest are composed of various components that interact with each other, often in a very complex manner. Spatial decomposition, in which components occupy distinct regions within the overall domain of the system, is one of the most common and important examples. Such a spatial decomposition may arise naturally (e.g., land, ocean, and atmosphere in a climate model) or may be artificially induced as part of a numerical discretization, especially when partitioned for parallel implementation. Interactions between adjacent components typically occur at the interfacial surfaces where their geometric subdomains abut. A further complication is that the problem geometry may change over time, and the interfacial surfaces between components may move. At the Center for Simulation of Advanced Rockets (CSAR) at the University of Illinois (www.csar.uiuc.edu), we have been developing a software system, Rocstar, for detailed simulation of solid rocket motors. This system involves many disciplines, including three broad physical disciplines—fluid dynamics, solid mechanics, and combustion—that interact with each other at the primary system level, with additional subsystem level interactions, such as particles and turbulence within fluids. To accommodate the diverse and dynamically changing needs of individual physics disciplines, we have adopted a partitioned approach, to enable coupling of individual software components that solve problems in their own physical and geometrical domains. With this approach, the physical components of the system are naturally mapped onto various software components (or modules), which can then be developed and parallelized independently. These modules are then integrated into a coherent system through an integration framework, which, among other responsibilities, manages distributed data objects and performs intermodule communications on parallel machines.