In this paper, we present a unimodular loop transformation called rotation as a simple, systematic and uniform method for partitioning the iteration spaces of doubly nested loops for execution on distributed memory multiprocessors. We define three parameters which could be used to choose an optimal rotation. These parameters are the parallelism factor, the !oad imbalance and the volume of communication. We present algebraic expressions for these parameters and discuss their relative significance in choosing a combined metric.
We present a computationally efficient method for deriving the most appropriate transformation and mapping of a nested loop for a given hierarchical parallel machine. This method is in the context of our systematic and general theory of unimodular loop transformations for the problem of iteration space partitioning [7]. Finding an optimal mapping or an optimal associated unimodular transformation is NP-complete. We present a polynomial time method for obtaining a 'good' transformation using a simple parameterized model of the hierarchical machine. We outline a systematic methodology for obtaining the most appropriate mapping.
Association rule mining (ARM) discovers correlations between different item sets in a transaction database. It provides important knowledge in business for decision makers. Association rule mining is an active data mining research area and most ARM algorithms cater to a centralized environment. Centralized data mining to discover useful patterns in distributed databases isn't always feasible because merging data sets from different sites incurs huge network communication costs. In this paper, an improved algorithm based on good performance level for data mining is being proposed. In local sites, it runs the application based on the improved LMatrix algorithm, which is used to calculate local support counts. Local Site also finds a center site to manage every message exchanged to obtain all globally frequent item sets. It also reduces the time of scan of partition database by using LMatrix which increases the performance of the algorithm. Therefore, the research is to develop a distributed algorithm for geographically distributed data sets that reduces communication costs, superior running efficiency, and stronger scalability than direct application of a sequential algorithm in distributed databases.
We present a unimodular transformation called rotation to partition the iteration space of a perfectly nested loop. The transformation captures the individual transformations like loop interchange, reversal, and skewing in a uniform framework. Though the trans formation is for any architecture, we have specifically addressed its application to distributed memory m* chines in this paper. We prove that the transformation is free from deadlocks. We finally describe the choice of an optimal rotation based on the pamllelism factor, load imbalance and the volume of communication in the transformed space.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.