A Distributed Algorithm for Balanced Hypergraph Partitioning

Hypergraph is good at modeling multi-node relationships in complex networks. Balanced hypergraph partitioning helps to optimize storage of large sets of hypergraph-structured data over multi-hosts in the Cloud, and share the query loads. Several centralized vertex partitioning algorithms have been developed to address this problem. However, edge partitioning is proved more effective than vertex partitioning for graph processing. Aim of this paper is to explore a new approach based on hyperedge partitioning, in which hyperedges, rather than vertices, are partitioned into disjoint subsets. We propose a distributed hyperedge partition algorithm, HyperSwap, to partition the hypergraph into balanced sub-hypergraph as required, without global information and central coordination. We show the feasibility, evaluate it on Facebook dataset with various settings, and compare it against two alternative solutions. Experiment findings show that HyperSwap outperforms the other two partitioners because it obtains good partitions with low cut cost while conforming to any balance requirement.