SSP 1995 project summary:

Large Matrix Operations in Large Scale Structure of the Universe

One of the central problems of Cosmology is developing a standard model, which will explain the large scale structure as a consequence of the physics of the early universe.

This project is part of a larger study which aims to analyse the large scale structure of the Universe. In order to model the large scale structure the manipulation of large non-sparse matrices is neccessary. The types of operations required to be performed on these matrices include inversion, diagonalisation, similarity transforms and Choleski decomposition. A realistic simulation of the problem would require the manipulation of matrices of up to 100,000 square elements, something which requires extremely large computational times.

The aim of this project is to investigate the performance of parallel versions of the matrix manipulations required in the large-scale structure study. The actual operations that will be examined here are the Choleski Decomposition, the Singular Value Decomposition and the Jacobi transformation. Different data distribution are looked at for each problem, while the actual algorithm for the solution of each problem is a parallel one. The results are timed and conclusions regarding the impact of the number of proccessors used and the size of the problem are made. The work is carried out using the Message Passing Interface (MPI) standard, so that the code will be portable to a number of different platforms.

Asimina Sierou worked on this project.

Compressed PostScript of the project's final report is available here (129148 bytes) .

Webpage maintained by mario@epcc.ed.ac.uk