SRConfig: An Empirical Method of Interdependent Soft Configurations for Improving Performance in n-Tier Application

Abstract

Efficient resources utilization and better system performance are always two important objectives that service providers pursue to enjoy a maximum profit. In this paper, through analyzing experimental measurements, we study the performance impact of interdependent soft resources on an n-tier application benchmark - the RUBiS system. Soft resources are vital factors that influence hardware resources usage and overall application performance. Improper soft configurations can result in correlated bottlenecks and make performance degradation, so tuning the configuration of soft resources is imperative. Based on the experimental measurements, SRConfig method is applied to predict the soft configurations through adopting the back propagation neural network in n-tier application. Experimental results validate the accuracy and efficacy of our method.