Configuration Optimization with Limited Functional Impact

Abstract

Dealing with a large configuration space is a complex task for developers, especially when configurations must comply with both functional constraints and non-functional goals. In this paper, we introduce an approach to optimize any set of performance indicators for an existing configuration, while meeting functional requirements. The efficiency of this approach is assessed by exhaustively optimizing a configurable system, and by analyzing how the algorithm navigates through the configuration space. This approach proves especially efficient at optimizing configurations through a minimal number of changes, thus limiting the impact on their functional behavior.