MosesA metaheuristic optimization software ecosystem. Applications to the automated analysis of software product lines and service-based applications

Résumé

In almost any area of human activity, the spirit of constant improvement and the re- quirements of performance, quality and efficiency, lead us to face scenarios of ever growing complexity. Most of these scenarios can be expressed as optimization prob- lems. Heuristic algorithms have been used for decades to guide the search of satisfy- ing solutions for hard optimization problems at an affordable execution time. In turn, Metaheuristics are reusable algorithm schemes that ease design heuristic algorithms. Even with metaheuristics, the process of designing and executing heuristic algo- rithms for solving optimization problems -so called the Metaheuristic Problem Solving (MPS) life-cycle in this thesis- is difficult and costly. The root of these difficulties is twofold: First, the design of algorithms using metaheuristics involves making deci- sions such as setting parameters values, choosing a solution encoding, etc. Since there does not exist an analytical method for making such decisions, a set of costly and de- manding experiments is required. Second, the MPS lifecycle involves implementing efficient programs which is non-trivial and error-prone. With the aim of reducing such complexity and costs of implementation, dozens of software frameworks -so called Metaheuristic Optimization Frameworks (MOFs)- have been proposed in literature. The goal of this thesis is to reduce the cost and complexity of executing the MPS lifecycle. To address this goal, a set of methods and tools is proposed. Specifically, the appropriate decision making in the MPS lifecycle is supported by: (i) a set of tools integrated into a software ecosystem that reduce the effort and time required by ex- perimentation; (ii) the automation of consistency validations on experimental design, conduction and analysis, decreasing the expertise required for experimentation; and (iii) increasing the degree of automation in experimental replication. Moreover, the se- lection of the appropriate MOF for a given problem is supported through a comparison framework and a survey. The proposed software ecosystem has been validated by solving two optimization problems in software engineering. It reduced the workload of executing the MPS life- cycle, lessening the implementation effort and the experimentation burden, and pro- ducing algorithms that improve the state of the art for both problems.