MAESTROS: multi-agent simulation of rework in Open Source software

• Autor(a) principal: Thiago R. P. M. Rúbio
• Data de Publicação: 2015
• Outros Autores: Henrique Lopes Cardoso, Eugénio da Costa Oliveira
• Tipo de documento: Livro
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://hdl.handle.net/10216/79998
• topic: Engenharia electrotécnica, electrónica e informática; Electrical engineering, Electronic engineering, Information engineering

Rework Management in software development is a challenging and complex issue. Defined as the effort spent to re-do some work, rework implies big costs given the fact that the time spent on rework does not count to the improvement of the project. Predicting and controlling rework causes is a valuable asset for companies, which maintain closed policies on choosing team members and assigning activities to developers. However, a trending growth in development consists in Open Source Software (OSS) projects. This is a totally new and diverse environment, in the sense that not only the projects but also their resources, e.g., developers change dynamically. There is no guarantee that developers will follow the same methodologies and quality policies as in a traditional and closed project. In such world, identifying rework causes is a necessary step to reduce project costs and to help project managers to better define their strategies. We observed that in real OSS projects there are no fixed team, but instead, developers assume some kind of auction in which the activities are assigned to the most interested and less-cost developer. This lead us to think that a more complex auctioning mechanism should not only model the task allocation problem, but also consider some other factors related to rework causes. By doing this, we could optimise the task allocation, improving the development of the project and reducing rework. In this paper we presented MAESTROS, a Multi-Agent System that implements an auction mechanism for simulating task allocation in OSS. Experiments were conducted to measure costs and rework with different project characteristics. We analysed the impact of introducing a Q-learning reinforcement algorithm on reducing costs and rework. Our findings correspond to a reduction of 31 % in costs and 11 % in rework when compared with the simple approach. Improvements to MAESTROS include real projects data analysis and a real-time mechanism to support Project Management decisions.