Background

Solving linear systems with sparse matrices is an important task in scientific computing. Use of parallel computers for such a task becomes mandatory when the problem size is large. However, different approaches to parallelization have different strengths. It is the purpose of the thesis work to study, in particular, how well the parallel algorithms that are implemented in SuperLU work: http://www.nersc.gov/~xiaoye/SuperLU/

What You Will Do

• Install and test the parallel version of SuperLU
• Do performance study on several parallel computers
• Do comparison with other approaches existing in Diffpack (http://www.diffpack.com)

What You Will Learn

• Parallel programming
• MPI
• C and C++

Qualifications

Required courses:

INF-MAT3350 or similar