Parallel Optimization of Frequent Algorithm on Multi-core Processors

Abstract

In this paper, we present a novel precision integrated framework (PRIF) to parallelize the famous Frequent algorithm in the context of multi-core processors. PRIF does this by equally distributing the stream of items into several sub-threads, each of which runs an optimized weighted Frequent algorithm independently and in parallel. The items with frequency increments over a pre-defined threshold in the sub-threads are sent to a merging thread which uses the same optimized weighted Frequent algorithm to provide the final e-deficient frequent items. The theoretical correctness analysis is presented. Experiments with three real traffic traces show that PRIF exhibits excellent scalability and delivers almost linear speedup.