Performance Comparison

Contents

Performance Comparison#

The executable ``linear-algebra-lapack`` contains our solver, which provides three solution methods accessible via the following flags: - -g: Gauss-Jordan Elimination - -c: Our primitive Cholesky solver - -cl: LAPACK Cholesky solver.

We aim to demonstrate that existing libraries often provide better performance than custom implementations. As expected, the well-optimized LAPACK library offers a much faster Cholesky method. To this end, we will solve a relatively large matrix system using the -c and -cl flags. The matrix used for this comparison is a 4515 x 4515 sparse, square matrix: Boeing/msc04515 . The matrix file, located at /scratch/vp91/msc04515.dat, has been sanitized from the original data to ensure it has the correct header for our solvers.

Exercise#

  1. Run the solver with the following two options. What metric(s) can you use to benchmark their performance?

./linear-algebra-lapack -c /scratch/vp91/msc04515.dat
./linear-algebra-lapack -cl /scratch/vp91/msc04515.dat

Note

Note that in practice, one might prefer using iterative methods to solve sparse systems such as msc04515.