An effectivemultileveltabusearchapproachforbalancedgraphpartitioning

Graph partitioningisoneofthefundamentalNP-completeproblemswhichiswidelyappliedinmany domains,suchasVLSIdesign,imagesegmentation,datamining,etc.Givenagraph G¼(V,E), thebalanced k-partitioningproblemconsistsinpartitioningthevertexset V into k disjointsubsetsofaboutthesame size, suchthatthenumberofcuttingedgesisminimized.Inthispaper,wepresentamultilevelalgorithm for balancedpartition,whichintegratesapowerfulrefinementprocedurebasedontabusearchwith periodicperturbations.Experimentalevaluationsonawidecollectionofbenchmarkgraphsshowthatthe proposedapproachnotonlycompetesveryfavorablywiththetwowell-knownpartitioningpackages METIS andCHACO,butalsoimprovesmorethantwothirdsofthebestbalancedpartitionseverreportedin the literature.