Data mining algorithms on the cell broadband engine

Abstract

The Cell Broadband Engine (CBE) is a new heterogeneous multi-core processor from IBM, Sony and Toshiba, and provides the potential to achieve an impressive level of performance for data mining algorithms. In this paper, we describe our implementation of three important classes of data mining algorithms: clustering (k-Means), classification (RBF network), and association rule mining (Apriori) on the CBE. We explain our parallelization methodology and describe the exploitation of thread- and data-level parallelism in each of the three algorithms. Finally we present experimental results on the Cell hardware, where we could achieve a high performance of up to 10 GFLOP/s and a speedup of up to 40. © 2008 Springer-Verlag Berlin Heidelberg.