next up previous contents index
Next: P_ARPACK Up: Introduction to ARPACK Previous: Dependence on LAPACK and

Expected Performance

  ARPACK has been designed for straightforward adaptation to a variety of high performance architectures including vector, super-scalar and parallel machines. It is intended to be portable and efficient across a wide range of computing platforms. Computationally intensive kernels are all expressed through BLAS operations and if the number k remains fixed as n increases the performance will scale asymptotically to the Level 2 BLAS operation _GEMV. Computational rates near maximum achievable peak are possible on multi-vector processors such as CRAY-C90 and on workstation clusters such as SGI Power Challenge.

The package is written in the ANSI standard Fortran 77 language with the one   exception of include files. These   are associated solely with the trace debugging facility provided with ARPACK. Each of the ARPACK subroutines reference two include files for debugging and timing purposes (see Appendix B). These references may be easily deleted if they are incompatible with your system.

Chao Yang