An Efficient Parallel Clustering Algorithm for Large Scale Database

摘要

In this paper, we propose a new parallel clusteringalgorithm, named Stem-Leaf-Point Plot ClusteringAlgorithm (SLPPCA). SLPPCA tends to produce clusters ofdifferent shapes and sizes, and according to our experiments,it can produces clusters more efficiently than traditionalmethods. SLPPCA can fully exploits the data-parallelism ofdata objects, and adopts a task decomposition design step tobalance the workloads of multi-core processors to achievea high speedup. We implemented SLPPCA to large scaledata base on duo-core processor and quad-core processorbased computer separately and analyzed its performance.The experimental results show that the clusters it producedwere particularly good either in different density or shapes,furthermore, with the parallel pattern used in SLPPCA onmulti-core platform, the speedup was almost linear withthe numbers of cores in processor and the number of datapoints. Moreover, SLPPCA can generate satisfactory clusternumber automatically in clustering process.