Many software systems provide configuration options for users to tailor their functional behavior as well as non-functional properties (e.g., performance, cost, and energy consumption). Configuration options relevant to users are often called features. Each variant derived from a configurable software system can be represented as a selection of features, called a configuration.
Performance (e.g., response time or throughput) is one of the most important non-functional properties, because it directly affects user perception and cost. To find an optimal configuration to meet a specific performance goal, it is crucial for developers and IT administrators to understand the correlation between feature selections and performance.
We investigate a practical approach that mines such a correlation from a sample of measured configurations, specifies the correlation as an explicit performance prediction model, and then uses the model to predict the performance of other unmeasured configurations.
Guo, J., K. Czarnecki, S. Apel, N. Siegmund, and A. Wasowski, “Variability-Aware Performance Prediction: A Statistical Learning Approach“, 28th IEEE/ACM International Conference on Automated Software Engineering (ASE), Silicon Valley, California, USA, IEEE, 11/2013.