Developing a Measure for Process Cluster Evaluation

Our preliminary research identi ed multiple approaches to automatic process clustering. Prototypical tools implementing these approaches were developed. The resulting cluster hierarchies have been evaluated by comparing them to the author's understanding of the application design. This paper discusses two quantitative measures for process cluster evaluation, taken from the literature and adopted to our speci c context. The rst set of measures, employing information about the interprocess communication during program execution, does not appear to be useful. The evaluation of clusters with these measures di ers from a human evaluation. The second measure, evaluating the degree of similarity among processes using information derived by a static source analysis, shows greater conformance with a human evaluation. This measure is improved by adding information about actual interprocess communication during runtime and an example is given that demonstrates the use of this measure.