|
Документ взят из кэша поисковой машины. Адрес
оригинального документа
: http://num-meth.srcc.msu.ru/english/zhurnal/tom_2014/v15r135.html
Дата изменения: Mon Jun 30 17:14:33 2014 Дата индексирования: Sun Apr 10 03:05:14 2016 Кодировка: IBM-866 |
|
"Optimization of a partitioning algorithm for a hypergraph with arbitrary weights
of vertices" Rusakov A.S. and Sheblaev M.V. |
|
One of the methods for the decomposition of a large problem to subproblems is its representation as a graph or hypergraph and partition this graph to approximately equal subgraphs with minimal cuts. The balanced hypergraph partitioning with the minimization of the cut size reduces communication cost between decomposed subproblems. The well-known approach to the hypergraph decomposition is the Fiduccia-Mattheyses (FM) algorithm and its hierarchical modifications. In this paper we discuss a key data structure modifications of the FM-algorithm to improve the performance and quality of the hierarchical partitioning algorithms and to reduce the computational overheads during solving large problems by parallel methods. Keywords: hypergraph partitioning, Fiduccia-Mattheyses algorithm, clustering, distributed computing systems, parallel programming.
|
|