cs 15(3): e3

Research Article

PerfCenterLite: Extrapolating Load Test Results for Performance Prediction of Multi-Tier Applications

Download275 downloads
  • @ARTICLE{10.4108/icst.valuetools.2014.258208,
        author={Varsha Apte and Nadeesh T. V.},
        title={PerfCenterLite: Extrapolating Load Test Results for Performance Prediction of Multi-Tier Applications},
        journal={EAI Endorsed Transactions on Cloud Systems},
        volume={1},
        number={3},
        publisher={EAI},
        journal_a={CS},
        year={2015},
        month={2},
        keywords={performance modeling, load test results},
        doi={10.4108/icst.valuetools.2014.258208}
    }
    
  • Varsha Apte
    Nadeesh T. V.
    Year: 2015
    PerfCenterLite: Extrapolating Load Test Results for Performance Prediction of Multi-Tier Applications
    CS
    EAI
    DOI: 10.4108/icst.valuetools.2014.258208
Varsha Apte1,*, Nadeesh T. V.2
  • 1: CSE Department, IIT Bombay
  • 2: CSE Department IIT Bombay
*Contact email: varsha@cse.iitb.ac.in

Abstract

Performance modeling is an important step in the lifecycle of a typical Web-based multi-tier application. However, while most practitioners are comfortable carrying out load tests on a Web application on a testbed, they find sophisticated performance modeling tools difficult to use because many inputs required by them are difficult to obtain. Chief among these is the service times of various types of requests at various resources in the multi-tier system (e.g. CPU execution time required at the Web server by a “Login” request). In this paper, we present PerfCenterLite, a tool focused on ease of use for practitioners of performance analysis. The tool (a) provides a spread-sheet template for describing the application architecture and (b) accepts standard performance metrics obtained from load testing of the application. PerfCenterLite then uses mathematical estimation techniques and transforms this input into a full-fledged performance model as required by a sophisticated performance modeling tool. Validation experiments show that performance metrics predicted using PerfCenterLite match well with measured values.