home search a-z help
University of Cambridge Computer Laboratory
Thursday Apr 6th, 2006 - 4:30pm
Computer Laboratory > Research > Systems Research Group > NetOS > Seminars > Thursday Apr 6th, 2006 - 4:30pm

A Practical Approach for Predicting the Performance and Scalability of Distributed Component Systems

Samuel Kounev

Distributed component systems (DCS) are becoming increasingly ubiquitous as enabling technology for modern enterprise applications. However, as systems grow in size and complexity, analyzing their expected performance and scalability becomes a more and more challenging task. System architects and deployers are often faced with questions such as the following:

  1. Which platform would provide the best cost/performance ratio for a given application?
  2. How do we ensure that the selected platform does not have any inherent scalability bottlenecks?
  3. What performance would the application exhibit and how much hardware would be needed to meet SLAs?
  4. How do we ensure that the application design does not have any inherent scalability bottlenecks?

Answering the first two questions requires measuring the performance and scalability of platforms for DCS. Answering the second two questions requires predicting the performance of a given system deployed on a selected platform. In this talk, we present a systematic approach for performance engineering of DCS that helps to address these questions. The proposed approach is based on a combination of benchmarking and performance modeling. We focus on J2EE-based platforms since they are currently the technology of choice for building DCS. In the first part of the talk, the major issues and challenges in benchmarking large-scale J2EE platforms are discussed considering several state-of-the-art benchmarks for J2EE. In the second part, a realistic performance modeling case study is presented in which a real-world J2EE application is modeled and its performance under load is predicted.