Monte Carlo Simulations for Variability Analyses in Highly Configurable Systems

Abstract

Highly configurable systems expose numerous variation points to be configured by the stakeholders. Deciding which variant to select for a given variation point is hard to know a priori because each variant affects the configuration properties (e.g., performance, efficiency, fault tolerance) differently, and evaluating all configurations is practically infeasible. This paper introduces an approach based on Monte Carlo simulations to analyze the influence of each feature selection when configuring a variability model.We split the whole configuration space into step-wise decisions driven by the variation points that a user normally needs to face/decide during the configuration process. Monte Carlo simulations approximate the influence of each feature variant evaluating as few configurations as possible. Our solution complements existing sampling techniques to analyze colossal configuration spaces improving the understanding of the influence of each feature selection. It can be part of a decisionmaking tool to assist the user by means of recommendation systems and interactive configuration processes.