Improved Feedback for Architectural Performance Prediction Using Software Cartography Visualizations

  • Authors:

  • Klaus Krogmann;Christian M. Schweda;Sabine Buckl;Michael Kuperberg;Anne Martens;Florian Matthes

  • Affiliations:

  • Software Design and Quality Group, Universität Karlsruhe (TH), Germany;Software Engineering for Business Information Systems, Technische Universität München, Germany;Software Engineering for Business Information Systems, Technische Universität München, Germany;Software Design and Quality Group, Universität Karlsruhe (TH), Germany;Software Design and Quality Group, Universität Karlsruhe (TH), Germany;Software Engineering for Business Information Systems, Technische Universität München, Germany

  • Venue:
  • QoSA '09 Proceedings of the 5th International Conference on the Quality of Software Architectures: Architectures for Adaptive Software Systems
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Software performance engineering provides techniques to analyze and predict the performance (e.g., response time or resource utilization) of software systems to avoid implementations with insufficient performance. These techniques operate on models of software, often at an architectural level, to enable early, design-time predictions for evaluating design alternatives. Current software performance engineering approaches allow the prediction of performance at design time, but often provide cryptic results (e.g., lengths of queues). These prediction results can be hardly mapped back to the software architecture by humans, making it hard to derive the right design decisions. In this paper, we integrate software cartography (a map technique) with software performance engineering to overcome the limited interpretability of raw performance prediction results. Our approach is based on model transformations and a general software visualization approach. It provides an intuitive mapping of prediction results to the software architecture which simplifies design decisions. We successfully evaluated our approach in a quasi experiment involving 41 participants by comparing the correctness of performance-improving design decisions and participants' time effort using our novel approach to an existing software performance visualization.