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.

Comparison of three algorithms for nonlinear metal cutting problems

This paper compares three alternative numerical algorithms applied to a nonlinear metal cutting problem. One algorithm is based on an explicit method and the other two are implicit. Domain decomposition (DD) is used to break the original domain into subdomains, each containing a properly connected, well-formulated and continuous subproblem. The serial version of the explicit algorithm is implemented in FORTRAN and its parallel version uses MPI (Message Passing Interface) calls. One implicit algorithm is implemented by coupling the state-of-the-art PETSc (Portable, Extensible Toolkit for Scientific Computation) software with in-house software in order to solve the subproblems. The second implicit algorithm is implemented completely within PETSc. PETSc uses MPI as the underlying communication library. Finally, a 2D example is used to test the algorithms and various comparisons are made.

Staggered-mesh computation for aerodynamic sound

For the numerical solution of the linearized Euler equations, an optimized computational scheme is considered. It is based on fully staggered (in space and time) regular meshes and on a simple mirroring procedure at the stepwise solid walls. There is no need to define ghost points into the solid ohjects that reflect the sound waves. Test results demonstrate the accuracy of the method that may be used for aeroacoustic problems with complex geometries.

A semi-Lagrangian finite volume method for Newtonian contraction flows

A new finite volume method for solving the incompressible Navier--Stokes equations is presented. The main features of this method are the location of the velocity components and pressure on different staggered grids and a semi-Lagrangian method for the treatment of convection. An interpolation procedure based on area-weighting is used for the convection part of the computation. The method is applied to flow through a constricted channel, and results are obtained for Reynolds numbers, based on half the flow rate, up to 1000. The behavior of the vortex in the salient corner is investigated qualitatively and quantitatively, and excellent agreement is found with the numerical results of Dennis and Smith [Proc. Roy. Soc. London A, 372 (1980), pp. 393-414] and the asymptotic theory of Smith [J. Fluid Mech., 90 (1979), pp. 725-754].

AC & DC magnetic levitation and semi-levitation modelling

This work presents computation analysis of levitated liquid thermal and flow fields with free surface oscillations in AC and DC magnetic fields. The volume electromagnetic force distribution is continuously updated with the shape and position change. The oscillation frequency spectra are analysed for droplets levitation against gravity in AC and DC magnetic fields at various combinations. For larger volume liquid metal confinement and melting the semi-levitation induction skull melting process is simulated with the same numerical model. Applications are aimed at pure electromagnetic material processing techniques and the material properties measurements in uncontaminated conditions.

The semi-automatic parallelisation of scientific application codes using a computer aided parallelisation toolkit

The shared-memory programming model can be an effective way to achieve parallelism on shared memory parallel computers. Historically however, the lack of a programming standard using directives and the limited scalability have affected its take-up. Recent advances in hardware and software technologies have resulted in improvements to both the performance of parallel programs with compiler directives and the issue of portability with the introduction of OpenMP. In this study, the Computer Aided Parallelisation Toolkit has been extended to automatically generate OpenMP-based parallel programs with nominal user assistance. We categorize the different loop types and show how efficient directives can be placed using the toolkit's in-depth interprocedural analysis. Examples are taken from the NAS parallel benchmarks and a number of real-world application codes. This demonstrates the great potential of using the toolkit to quickly parallelise serial programs as well as the good performance achievable on up to 300 processors for hybrid message passing-directive parallelisations.

A blackboard architecture for a hybrid CBR system for scientific software

This paper describes the use of a blackboard architecture for building a hybrid case based reasoning (CBR) system. The Smartfire fire field modelling package has been built using this architecture and includes a CBR component. It allows the integration into the system of qualitative spatial reasoning knowledge from domain experts. The system can be used for the automatic set-up of fire field models. This enables fire safety practitioners who are not expert in modelling techniques to use a fire modelling tool. The paper discusses the integrating powers of the architecture, which is based on a common knowledge representation comprising a metric diagram and place vocabulary and mechanisms for adaptation and conflict resolution built on the Blackboard.

Predicting the behaviour of micro-electronic displays using computational mechanics

Micro-electronic displays are indispensible devices used in high performance applications such as aerospace, medical, marine and industrial sectors.These devices provide an interface to real time mission critical devices and therefore require good optical visual performance and high reliability, all this within varied and challenging environments.

Using an interactive software environment for the parallelization of real-world scientific applications

The parallelization of real-world compute intensive Fortran application codes is generally not a trivial task. If the time to complete the parallelization is to be significantly reduced then an environment is needed that will assist the programmer in the various tasks of code parallelization. In this paper the authors present a code parallelization environment where a number of tools that address the main tasks such as code parallelization, debugging and optimization are available. The ParaWise and CAPO parallelization tools are discussed which enable the near automatic parallelization of real-world scientific application codes for shared and distributed memory-based parallel systems. As user involvement in the parallelization process can introduce errors, a relative debugging tool (P2d2) is also available and can be used to perform nearly automatic relative debugging of a program that has been parallelized using the tools. A high quality interprocedural dependence analysis as well as user-tool interaction are also highlighted and are vital to the generation of efficient parallel code and in the optimization of the backtracking and speculation process used in relative debugging. Results of benchmark and real-world application codes parallelized are presented and show the benefits of using the environment