Scalable Tools for Scientific Simulation

Linear and non-linear solvers, discretizations, load balancing, distributed data structures and efficient parallel communication are among the critical components of scalable parallel scientific simulations.  Many software projects exist to provide these capabilities and ease the development of scalable applications.  In this tutorial, we describe Sandia National Laboratories’ Trilinos and Zoltan projects.  Trilinos is a collection of software packages providing linear solvers, non-linear solvers, eigen-solvers, preconditioners, automatic differentiation, matrix/vector classes, and matrix partitioning within a framework of shared parallel matrix/vector interfaces and operations.  Zoltan is a toolkit of partitioning and dynamic load-balancing algorithms for a wide range of parallel, unstructured and/or dynamic applications.  In this tutorial, we include an overview of both projects, as well as details on how they can be exploited separately or together to develop scalable applications.

Close window