Numerical analyses of aero-optical flows: An application of parallel algorithm on high-performance personal computer (PC) base multitransputer system

Numerical analyses of compressible viscous flow using a {dollar}p{dollar}ersonal {dollar}c{dollar}omputer ({dollar}PC{dollar}) with a single {dollar}transputer{dollar} ({dollar}trans{dollar}istor and comp{dollar}uter{dollar}) and a multi-transputer system are presented here. Two objectives are achieved: (1) Basic features of the aero-optical flows is investigated by using the multi-dimensional {dollar}U{dollar}pwind {dollar}F{dollar}lux {dollar}D{dollar}ifference {dollar}S{dollar}plitting ({dollar}UFDS{dollar}) implicit algorithm to solve the full compressible Navier-Stokes equations. The algorithm is validated and is in good agreement with experiment data. Results confirm that this algorithm has the capability to solve a wide range of fluid problems. Predictions generated by the code replicate the same flowfield features for the complex flow phenomenon which aero-optical interactions will encounter. (2) A parallel algorithm of the present {dollar}CFD{dollar} code is implemented for a {dollar}PC{dollar}-base transputer system. A single transputer version and a four-transputer version of transputer {dollar}CFD{dollar} code are developed. The single transputer {dollar}CFD{dollar} version is used to compare the significance of turnaround time and machine roundoff errors with four different computers. Comparisons of four-transputer {dollar}CFD{dollar} results and single transputer {dollar}CFD{dollar} results are also presented.; Solutions generated by the transputer do not show any significant roundoff errors. The performance of multi-processing is below expectations due to a large overhead of synchronization and communication between processors. However, this performance can be improved significantly by some hardware and software modifications. As a whole, the transputer system demonstrates that it is feasible and more economical for {dollar}CFD{dollar} computations.