Software bottleneck analysis during performance testing

Abstract

Scalability of a multi-tier enterprise systemis limited resources that becomes a bottleneck, by the presence of software and hardware resource bottlenecks. These bottlenecks typically occur at larger number of users. From an IT industry point of view, deployment process of enterprise applications becomes simpler if these bottlenecks are known apriori during the performance testing itself. This paper uses an analytical model based technique for analyzing performance of such a system where the model consists of two layers of queuing networks for software resources and hardware resources. Proposed solution strategy involves identifying all the software resources in the application, estimating their service demands along with service demands of hardware resources, incorporating these parameters into the model and finally solving it. The paper describes a methodology that uses these steps to identify software and hardware bottlenecks for a given enterprise application. The paper further presents two case studies dealing with real-life multi-tier enterprise applications that encounter software resource bottlenecks. The case studies show that the model is able to predict throughput and utilization of servers with accuracy close to 90%.