Three parallel programming paradigms: Comparisons on an archetypal PDE computation

Three paradigms for distributed-memory parallel computation that free the application programmer from the details of message passing are compared for an archetypal structured scientific computation -- a nonlinear, structured-grid partial differential equation boundary value problem -- using the same algorithm on the same hardware. All of the paradigms -- parallel languages represented by the Portland Group's HPF, (semi-)automated serial-to-parallel source-to-source translation represented by CAP-Tools from the University of Greenwich, and parallel libraries represented by Argonne's PETSc -- are found to be easy to use for this problem class, and all are reasonably effective in exploiting concurrency after a short learning curve. The level of involvement required by the application programmer under any paradigm includes specification of the data partitioning, corresponding to a geometrically simple decomposition of the domain of the PDE. Programming in SPMD style for the PETSc library requires writing only the routines that discretize the PDE and its Jacobian, managing subdomain-to-processor mappings (affine global-to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implementation of the same algorithm as a starting point, introduction of concurrency through subdomain blocking (a task similar to the index mapping), and modest experimentation with rewriting loops to elucidate to the compiler the latent concurrency. Programming with CAPTools involves feeding the same sequential implementation to the CAPTools interactive parallelization system, and guiding the source-to-source code transformation by responding to various queries about quantities knowable only at runtime. Results representative of "the state of the practice" for a scaled sequence of structured grid problems are given on three of the most important contemporary high-performance platforms: the IBM SP, the SGI Origin 2000, and the CRAYY T3E.

Assessing the scalability of multiphysics tools for modeling solidification and melting processes on parallel clusters

A comprehensive solution of solidification/melting processes requires the simultaneous representation of free surface fluid flow, heat transfer, phase change, nonlinear solid mechanics and, possibly, electromagnetics together with their interactions, in what is now known as multiphysics simulation. Such simulations are computationally intensive and the implementation of solution strategies for multiphysics calculations must embed their effective parallelization. For some years, together with our collaborators, we have been involved in the development of numerical software tools for multiphysics modeling on parallel cluster systems. This research has involved a combination of algorithmic procedures, parallel strategies and tools, plus the design of a computational modeling software environment and its deployment in a range of real world applications. One output from this research is the three-dimensional parallel multiphysics code, PHYSICA. In this paper we report on an assessment of its parallel scalability on a range of increasingly complex models drawn from actual industrial problems, on three contemporary parallel cluster systems.

Combined vertex-based - cell-centred finite volume method for flows in complex geometries

CFD modelling of 'real-life' processes often requires solutions in complex three dimensional geometries, which can often result in meshes where aspects of it are badly distorted. Cell-centred finite volume methods, typical of most commercial CFD tools, are computationally efficient, but can lead to convergence problems on meshes which feature cells with high non-orthogonal shapes. The vertex-based finite volume method handles distorted meshes with relative ease, but is computationally expensive. A combined vertex-based - cell-centred (VB-CC) technique, detailed in this paper, allows solutions on distorted meshes that defeat purely cell-centred physical models to be employed in the solution of other transported quantities. The VB-CC method is validated with benchmark solutions for thermally driven flow and turbulent flow. An early application of this hybrid technique is to three-dimensional flow over an aircraft wing, although it is planned to use it in a wide variety of processing applications in the future.

Predicting HCl concentrations in fire enclosures using an HCl decay model coupled to a CFD-based fire field model

The amount of atmospheric hydrogen chloride (HCl) within fire enclosures produced from the combustion of chloride-based materials tends to decay as the fire effluent is transported through the enclosure due to mixing with fresh air and absorption by solids. This paper describes an HCl decay model, typically used in zone models, which has been modified and applied to a computational fluid dynamics (CFD)-based fire field model. While the modified model still makes use of some empirical formulations to represent the deposition mechanisms, these have been reduced from the original three to two through the use of the CFD framework. Furthermore, the effect of HCl flow to the wall surfaces on the time to reach equilibrium between HCl in the boundary layer and on wall surfaces is addressed by the modified model. Simulation results using the modified HCl decay model are compared with data from three experiments. The model is found to be able to reproduce the experimental trends and the predicted HCl levels are in good agreement with measured values

A hybrid Laplace transform/finite difference boundary element method for diffusion problems

The solution process for diffusion problems usually involves the time development separately from the space solution. A finite difference algorithm in time requires a sequential time development in which all previous values must be determined prior to the current value. The Stehfest Laplace transform algorithm, however, allows time solutions without the knowledge of prior values. It is of interest to be able to develop a time-domain decomposition suitable for implementation in a parallel environment. One such possibility is to use the Laplace transform to develop coarse-grained solutions which act as the initial values for a set of fine-grained solutions. The independence of the Laplace transform solutions means that we do indeed have a time-domain decomposition process. Any suitable time solver can be used for the fine-grained solution. To illustrate the technique we shall use an Euler solver in time together with the dual reciprocity boundary element method for the space solution