Section: New Software and Platforms

SCOTCH

Keywords: Mesh partitioning - Domain decomposition - Graph algorithmics - High-performance calculation - Sparse matrix ordering

Functional Description: Scotch is a graph partitioner. It helps optimise the division of a problem, by means of a graph, into a set of independent sub-problems of equivalent sizes. These sub-problems can also be solved at the same time.

Release Functional Description: Version 6.0 offers many new features:

sequential graph repartitioning

sequential graph partitioning with fixed vertices

sequential graph repartitioning with fixed vertices

new, fast, direct k-way partitioning and mapping algorithms

multi-threaded, shared memory algorithms in the (formerly) sequential part of the library

exposure in the API of many distributed graph handling routines

embedded pseudo-random generator for improved reproducibility

and even more...

News Of The Year: In the context of the PhD of Rémi Barat, the sequential version of Scotch has been extended so as to manage graphs with multiple vertex weights, and multi-constraint graph partitioning algorithms have been implemented as prototypes.

• Participants: Sébastien Fourestier, François Pellegrini and Cédric Chevalier

• Partners: CNRS - IPB - Region Aquitaine

• Contact: François Pellegrini

• Publications: Process Mapping onto Complex Architectures and Partitions Thereof - Contributions au partitionnement de graphes parallèle multi-niveaux - Adaptation au repartitionnement de graphes d'une méthode d'optimisation globale par diffusion - A parallelisable multi-level banded diffusion scheme for computing balanced partitions with smooth boundaries - PT-Scotch: A tool for efficient parallel graph ordering - Design and implementation of efficient tools for parallel partitioning and distribution of very large numerical problems - PT-Scotch : Un outil pour la renumérotation parallèle efficace de grands graphes dans un contexte multi-niveaux - PT-Scotch: A tool for efficient parallel graph ordering - Improvement of the Efficiency of Genetic Algorithms for Scalable Parallel Graph Partitioning in a Multi-Level Framework

• URL: http://www.labri.fr/~pelegrin/scotch/