Towards a Taxonomy for the Evaluation of Self-* Software