DSCPACK can be used to solve sparse linear systems using direct methods
on multiprocessors and networks-of-workstations. This package is
suitable for systems where the coefficient matrix
is symmetric and sparse.
This solver is written in C; it uses MPI for
inter-processor communication and the BLAS
library for improved cache-performance.
Interfaces to DSCPACK are available through well-established
problem solving environments including
PETSc Toolkit for Scientific Computing from the Argonne National Laboratory,
Trilinos Software from Sandia National Laboratories.
Dscpack has been distributed to over 50 sites since 2000 and
is in active use in several projects including a crack-modeling
project at NASA and the Salinas project at Sandia National Laboratories.
Reference: DSCPACK: Domain-Separator Codes for the parallel solution of sparse linear
systems, P. Raghavan, Tech. Rep. CSE-02-004, Department of Computer Science and
Engineering, The Pennsylvania State University, University Park, PA
Postscript version at:
PDF version at:
Author: P. Raghavan
Please complete the short form at
http://www.cse.psu.edu/~raghavan/Dscpack/form.html to receive
a userid and password to access the software.
Click here to complete and submit the software request form.
Download the software at:
A Parallel Direct-Iterative Hybrid Solver
Through Flexible Incomplete Cholesky Preconditioning
Author: K. Teranishi
This solver provides scalable hybrids using incomplete Cholesky with thresholding
(ICT) with selective inversion (ICT-SI) or
selective sparse approximate inversion (ICT-SSAI) to
allow latency tolerant parallel application of the preconditioner.
Click here for the performamce of
these schemes on augustus7 (dimension over 1 million).
Click here for a presentation describing
ICT-SI and ICT-SSAI hybrid solvers.
This software is based on new schemes developed in
A test release is available at: