Inducing Load Balancing and Efficient Data Distribution Prior to Association Rule Discovery in a Parallel Environment

Many association rule algorithms operate in a parallel environment where the database is divided up among a number of processors, a procedure which is usually carried out indiscriminately. The nature of the database partitioning can affect both the number of candidate sets produced and the workload at each processor. This paper demonstrates that Principal Component Analysis can be used successfully to help arrange the records of a database among processors so that efficient load balancing is enabled and candidate set duplication minimised.