Detail publikace

Parallel Linear Equations Solvers for Scientific Simulation: Cluster and SMP Experience

Originální název

Parallel Linear Equations Solvers for Scientific Simulation: Cluster and SMP Experience

Anglický název

Parallel Linear Equations Solvers for Scientific Simulation: Cluster and SMP Experience

Jazyk

en

Originální abstrakt

Even though solving linear systems of equations Ax = b has been studied for long time, one cannot depend on numerical software libraries entirely due to a large number of all possible structures of A and due to many recent changes in high performance computer architectures, programming languages and compilers. If the user's priority is maximum performance on one specialized problem, custom software should be preferred. Here we describe experience with parallel solution of large linear systems of equations on shared as well as distributed memory architecture. The results of exact or iterative solution have been generated in Practical parallel programming course by student teams as term projects. The methods of solutions span from Gauss elimination to Gauss-Seidel or SOR iterations. Speedup and efficiency of various exact and approximate (iterative) methods of solution are compared on both shared memory as well as distributed memory architectures (clusters).

Anglický abstrakt

Even though solving linear systems of equations Ax = b has been studied for long time, one cannot depend on numerical software libraries entirely due to a large number of all possible structures of A and due to many recent changes in high performance computer architectures, programming languages and compilers. If the user's priority is maximum performance on one specialized problem, custom software should be preferred. Here we describe experience with parallel solution of large linear systems of equations on shared as well as distributed memory architecture. The results of exact or iterative solution have been generated in Practical parallel programming course by student teams as term projects. The methods of solutions span from Gauss elimination to Gauss-Seidel or SOR iterations. Speedup and efficiency of various exact and approximate (iterative) methods of solution are compared on both shared memory as well as distributed memory architectures (clusters).

BibTex


@inproceedings{BUT10044,
  author="Jiří {Staroba} and Václav {Dvořák}",
  title="Parallel Linear Equations Solvers for Scientific Simulation: Cluster and SMP Experience",
  annote="Even though solving linear systems of equations Ax = b has been studied for long time, one cannot depend on numerical software libraries entirely due to a large number of all possible structures of A and due to many recent changes in high performance computer architectures, programming languages and compilers. If the user's priority is maximum performance on one specialized problem, custom software should be preferred. Here we describe experience with parallel solution of large linear systems of equations on shared as well as distributed memory architecture. The results of exact or iterative solution have been generated in Practical parallel programming course by student teams as term projects. The methods of solutions span from Gauss elimination to Gauss-Seidel or SOR iterations. Speedup and efficiency of various exact and approximate (iterative) methods of solution are compared on both shared memory as well as distributed memory architectures (clusters).",
  booktitle="Proceedings of XXIVth International Autumn Colloquium ASIS'02 Advanced Simulation of Systems",
  chapter="10044",
  year="2002",
  month="september",
  pages="225--230",
  type="conference paper"
}