A Performance Evaluation of QR-eigensolver on IBM Roadrunner cluster for Large Sparse Matrices

Abstract

The paper presents a performance analysis of theQR eigensolver from ScaLAPACK library on the IBMRoadrunner machine. A ScaLAPACK-based testing platformwas developed in order to evaluate the performance of a parallelsolver to compute the eigenvalues and eigenvectors for largescalesparse matrices. Our experiments showed encouragingresults on the IBM Roadrunner cluster, the acceleration factorgained was up to 40 for large matrices. This result is bright tosolve problems that involve scientific and large-scale computing.

Keywords

• eigenvalues
• high performance computing
• QR factorization
• systems of linear equations
• ScaLAPACK