Font Size: a A A

Implementation and performance of a particle simulation method suited to MIMD parallel computer architectures

Posted on:1995-08-13Degree:Ph.DType:Dissertation
University:Stanford UniversityCandidate:Fallavollita, Michael AlfredFull Text:PDF
GTID:1478390014990424Subject:Engineering
Abstract/Summary:
Particle methods are useful for a wide range of problems of current interest to the scientific community. In addition to the usual application to hypersonic rarefied flows, particle methods can also be useful for modeling other flows in the rarefied regime such as the forces on the read/write head of a modern disk drive. The major drawback of particle simulation methods is the large amount of computer resources that they require, both in terms of processor time and memory. The current trend for supercomputer architectures toward large-scale MIMD (multiple-instruction multiple-data) parallel computers has provided an impetus for implementing a particle method which is well suited to these architectures.; A simulation code has been developed for the Intel Gamma and Delta computers which is capable of computing three-dimensional flow fields in multiple-species gasses that exhibit thermo-chemical non-equilibrium. Details of the implementation are presented that focus on those issues which arise from the parallel nature of the computers. Important aspects of the Intel Gamma and Delta architectures are presented in as much as they affect the implementation decisions. A very thorough study of the performance of the method is also presented which relates the efficiency of the implementation to the number of processors that are used. This is very important in order to predict the usefulness of this method, or similar methods on future computers which contain more processors. The study was also very useful for identifying those areas of the simulation which required additional optimization.; Finally, some results are presented which demonstrate the capabilities of the method. A study on how the number of particles in a simulation affect the simulation cost and accuracy shows that the computational cost of a simulation of a given accuracy may actually be reduced by increasing the number of particles. A comparison between particle simulation results and the solution of the Burnett equations is made for supersonic flow over a circular cylinder, and three-dimensional flow calculations about a blunted cone at varying angles of attack are also presented.
Keywords/Search Tags:Particle, Method, Simulation, Implementation, Presented, Parallel, Architectures
Related items