Parallel Implementation of SPARSKIT Using MPI on a Beowulf Cluster

Presenter Information

Joseph Dobmeier, Minnesota State University, Mankato

Location

CSU 253/4/5

Start Date

5-4-2010 10:00 AM

End Date

5-4-2010 12:00 PM

Student's Major

Mechanical and Civil Engineering

Student's College

Science, Engineering and Technology

Mentor's Name

Dean Kelley

Mentor's Department

Computer Information Science

Mentor's College

Science, Engineering and Technology

Second Mentor's Name

Steven Case

Second Mentor's Department

Computer Information Science

Second Mentor's College

Science, Engineering and Technology

Description

The finite element method (FEM) is an approximation technique used to simulate physical systems in many areas of science and engineering. A solution of high accuracy can be obtained by using many elements, at the cost of increased computation time. To mitigate this cost, the computation can be distributed across multiple processing units. The purpose of this research was to parallelize a FORTRAN library of iterative solver routines for FEM matrices called SPARSKIT using the message passing interface (MPI) protocol. Resulting computation times were then analyzed for speedup relative to the single processor solution. All computation was performed on the 140 processor Beowulf cluster at MSU.