A performance comparison study of parallel clustering algorithms in cluster environments

We compare the performance of three parallel clustering algorithms: Canopy, K-means and fuzzy K-means in real cluster environments. By constructing cluster platform of different scale, we compare these algorithms from three metrics: run time, speedup and sizeup. Experimental results show that: (1) if both the data set and the number of nodes in the cluster are the same, both the runtime and the sizeup of Canopy is the smallest, and the speedup of Canopy is the largest among these three algorithms; (2) when the number of nodes is larger than 2, for each algorithm, the difference between the sizeup on different data set becomes more and more smaller as the number of nodes increases; at the same time, when the number of nodes is larger than 2 and the data set is the same, the sizeup of Canopy has the maximum decrease with the increase of the number of nodes; (3) when the number of nodes keeps the same, the difference between the sizeup of these three algorithms becomes more and more apparent as the size of the data set increases.