Engineering Initial Partitioning Algorithms for direct k-way Hypergraph partitioning
نویسندگان
چکیده
Hypergraphs are generalizations of graphs where an edge can consist of more than two nodes. A reoccurring task is to divide the node set of a hypergraph into k different non-empty parts where we simultaneously want to minimize a partitioning objective. This problem is called the k-way hypergraph partitioning problem. Applications can be found in the area of VLSI design and the parallelization of computationally intensive problems. A method to solve this problem is the multilevel partitioning scheme. This scheme consist of three phases. The hypergraph is first coarsened, then initially partitioned and at the end the quality of the partition is improved with a local search heuristic. A special case of this scheme is the n-Level hypergraph partitioning framework KaHyPar. This framework currently uses the external tools hMetis and PaToH for the initial partitioning phase. In this thesis we develop various initial partitioning methods with the goal to produce the same quality in the same amount of time as hMetis in this framework. Our final initial partitioner combines all developed methods into one single initial partitioning algorithm. The solution quality of KaHyPar with our initial partitioner is comparable to the quality with hMetis and 1% better as the quality with PaToH as initial partitioner. The initial partitions produced by our initial partitioning algorithm are 0.6% better than the partitions of hMetis and the running time is 16% faster on average. Zusammenfassung Hypergraphen sind Generalisierungen von Graphen bei denen eine Kante aus mehr als nur aus zwei Knoten bestehen kann. Eine immer wiederkehrende Aufgabe ist es die Knotenmenge eines Hypergraphen in k verschiedene nicht-leere Teile aufzuteilen, wo wir gleichzeitig versuchen eine Partitionierungszielfunktion zu minimieren. Dieses Problem wird direktes k-way Hypergraph Partitionierungsproblem genannt. Einige Anwendungen sind im Bereich des VLSI -Design und bei der Parallelisierung von rechenaufwendigen Problemen zu finden. Eine Methode dieses Problem zu lösen ist das Multilevel-Partitionierungsschema. Dieses Verfahren besteht aus drei Phasen. Der Hypergraph wird als erstes vergröbert, danach initial partitioniert und am Ende wird eine lokale Suchheuristik dazu genutzt um die Partitionierung zu verbessern. Eine Speziallisierung dieses Schema is das n-level Hypergraph Partitionierungsframework KaHyPar. Dieses Framework benutzt im Moment die externen Anwendungen hMetis und PaToH für den initialen Partitionierungsschritt. In dieser Arbeit entwickelten wir verschiedene initiale Partitionierungsalgorithmen mit dem Ziel die gleiche Qualität in der gleichen Laufzeit im Vergleich zu hMetis in diesem Framework zu produzieren. Der letztendlich beste initiale Partitionierer kombiniert alle entwickelten Methoden in einem einzigen Algorithmus. Die Qualität der Lösungen von KaHyPar mit unserem initialen Partitionierer ist vergleichbar mit der Qualität mit hMetis und 1% besser als die Qualität mit PaToH als initialem Partitionierer. Die initialen Partitionen die von unserem initialen Partitionierungsalgorithmus produziert werden sind 0.6% besser als die Partitionen von hMetis und die Laufzeit ist im Durchschnitt 16% schneller.
منابع مشابه
Multi-level direct K-way hypergraph partitioning with multiple constraints and fixed vertices
K-way hypergraph partitioning has an ever-growing use in parallelization of scientific computing applications. We claim that hypergraph partitioning with multiple constraints and fixed vertices should be implemented using direct K-way refinement, instead of the widely adopted recursive bisection paradigm. Our arguments are based on the fact that recursive-bisection-based partitioning algorithms...
متن کاملMulti - level direct K - way hypergraph partitioning with multiple constraints and fixed vertices 1
K-way hypergraph partitioning has an ever-growing use in parallelization of scientific computing applications. We claim that hypergraph partitioning with multiple constraints and fixed vertices should be implemented using direct K-way refinement, instead of the widely adopted recursive bisection paradigm. Our arguments are based on the fact that recursive-bisection-based partitioning algorithms...
متن کاملLabel Propagation for Hypergraph Partitioning
Many problems in computer science can be represented by a graph and reducedto a graph clustering or k-way partitioning problem. In the classical definition,a graph consists of nodes and edges which usually connect exactly two nodes.Hypergraphs are a generalization of graphs, where every edge can connect anarbitrary number of nodes. Recent results suggest that some problems i...
متن کاملHypergraph Partitioning and Clustering
A hypergraph is a generalization of a graph wherein edges can connect more than two vertices and are called hyperedges. Just as graphs naturally represent many kinds of information in mathematical and computer science problems, hypergraphs also arise naturally in important practical problems, including circuit layout, Boolean SATisfiability, numerical linear algebra, etc. Given a hypergraph H ,...
متن کاملConstrained Min-Cut Replication for K-Way Hypergraph Partitioning
R is a widely-used technique in information retrieval and database systems for providing fault tolerance and reducing parallelization and processing costs. Combinatorial models based on hypergraph partitioning are proposed for various problems arising in information retrieval and database systems. We consider the possibility of using vertex replication to improve the quality of hypergraph parti...
متن کامل