Large software systems are often designed with limited explicit consideration of the implementation’s performance (e.g., response times, resource utilization). This ‘fix-performance-later’ attitude can lead to expensive redesigns in late software development stages. Model-based performance prediction methods try to assess the performance properties of systems already during early development stages. They rely on well-known mathematical formalisms, such as queueing networks, stochastic Petri nets or stochastic process algebra. Performance prediction based on numerical analysis or simulation can reveal potential bottlenecks in a software design and helps software architects to improve architectural designs.

Introductory papers to research on performance engineering:

• Smith et al.: Performance evaluation of software architectures (1998)
• Balsamo et al.: Model-based performance prediction in software development: A survey (2004)
• Woodside et al.: Performance by unified analysis (2005)
• Woodside et al: The future of software performance engineering (2007)

My PhD thesis focused on performance prediction for component-based software systems. I developed a parameterizable performance specification formalism for software components, which was later incorporated into the Palladio Component Model (PCM). Recent works originate from Anne’s PhD thesis on automatically improving architectural models for performance using evolutionary algorithms.

Selected papers with my participation:

• Performance prediction with different usage profiles
  (QoSA-Paper, 2007, LNCS)
• Performance prediction with the Palladio Component Model
  (JSS-Paper, 2009, workshop version won ACM Best Paper Award)
• A Survey of Performance Evaluation Methods for Component-based Software
  (Journal on Performance Evaluation, 2010, Elsevier)
• Automatically Improving Architecture Models for Performance
  (ICPE 2010, ACM)
•  An industrial case study of performance and cost design space exploration
  (ICPE 2012, ACM, won ACM Best Paper Award)