An Improved Hypergraph Partitioning Model for Parallel Scientific Computing

K-way hyper graph partitioning has an increasing use in parallel scientific computing because it can accurately model communication volume and has more expressions. However, the main shortcoming of hyper graph partitioning is that minimizing the so-called hyper edge cut is not entirely the same as minimizing the communication overhead, this is because it does not include the effects of communication latency and the distribution of communication overhead. We thus propose an improved hyper graph partitioning model that can take into account all these factors. Moreover, freely adjustable weighting parameters in the model also promote a flexible treatment of different optimization objectives. We also give a small scale hyper graph to verify the validity of the proposed model.