Performance Analysis of Parallel Software for Solving Systems of Linear Equations

Abstract

This paper aims to present and analyze the performance of various software solving systems of linear equations. Such systems can be seen as dense or sparse matrices. For software like SuperLU there is a need to address the problem at hand in a special way, developing an algorithm to parallel systems that make use of distributed memory and need to work with a specific interface, such as MPI. For software like Pardiso there is no need for data parallel approach of the algorithm, because the software takes care of itself, making the parallelism of areas where this situation is advantageous, but it is not possible to verify what is actually happening with the code in terms of the division of tasks and resources.