On Modeling Dependency between MapReduce Configuration Parameters and Total Execution Time

Abstract

In this paper, we propose an analytical method to model the dependency between configuration parameters and total execution time of Map-Reduce applications. Our approach has three key phases: profiling, modeling, and prediction. In profiling, an application is run several times with different sets of MapReduce configuration parameters to profile the execution time of the application on a given platform. Then in modeling, the relation between these parameters and total execution time is modeled by multivariate linear regression. Among the possible configuration parameters, two main parameters have been used in this study: the number of Mappers, and the number of Reducers. For evaluation, two standard applications (WordCount, and Exim Mainlog parsing) are utilized to evaluate our technique on a 4-node MapReduce platform.