Skip to Main Content
In this paper, we present parallel algorithms for Web log mining and the performance prediction model. The algorithm, based on WAP-tree, scans dataset only twice and avoids candidate generation process. We parallelized mining part of WAP tree. To balance the workload among processors, we developed a task scheduling strategy. A performance model of parallel Web mining algorithm is also developed to predict the performance of parallel implementation. This model shows that we can get linear speedup for a small number of processors, and a slow down of speedup as the number of processors increases. Using the performance model, we can also estimate the maximum speed up. We implemented the algorithm on a Pittsburg Super Computer Center Lemieux using up to 48 processors. Our benchmark results showed that the performance model correctly predicts the performance of the parallel implementation. We have achieved a good speedup as the size of the dataset is increased.
Date of Conference: 7-9 July 2004