An unbalanced partitioning scheme for graph in heterogeneous computing

Authors:
Yiwei Shen;Guosun Zeng
Affiliations:
Department of computer Science and Technology, Tongji University, Shanghai, China;Department of computer Science and Technology, Tongji University, Shanghai, China
Venue:
GCC'05 Proceedings of the 4th international conference on Grid and Cooperative Computing
Year:
2005

Citing 5
Cited 0

A unified geometric approach to graph separators

SFCS '91 Proceedings of the 32nd annual symposium on Foundations of computer science
An improved spectral graph partitioning algorithm for mapping parallel computations

SIAM Journal on Scientific Computing
Multilevel k-way partitioning scheme for irregular graphs

Journal of Parallel and Distributed Computing
A Fast and High Quality Multilevel Scheme for Partitioning Irregular Graphs

SIAM Journal on Scientific Computing
Guest Editor's Introduction: Heterogeneous Processing

Computer

Quantified Score

Hi-index	0.00

Visualization

Abstract

Most existing graph partitioning algorithms produce equivalent partitions. It means the partitioned subsets have equal number of vertexes. However, in heterogeneous computing, the computing power of different processors varies, so that the size of the tasks to be scheduled should not be the same as well. In order to meet the need of the load balance for heterogeneous computing, we present a novel algorithm that partitions the original task graph into unbalanced subsets according to the arbitrarily given conditions. Generally, the number of the partitions is equal to that of the processors and the size of each partition is set according to the computing power. Our algorithm contains three phases. Coarsen the original graph, then partition the coarsest graph, and finally project it back to the original graph and conduct refinement. We test our algorithm using Greenwich Graph Partitioning Archive and get good experimental results.