Evaluating process clusters to support automatic program understanding

Evaluating the design of a distributed application is difficult but provides useful information for program development and maintenance. In distributed debugging, for example, processes are often grouped together and treated as one entity to reduce the debugging complexity. We previously identified multiple approaches to automatic process clustering and prototypical tools implementing these approaches have been developed. The process clusters derived with these tools have been evaluated by comparing them to the author´s understanding of the application design. This paper discusses a quantitative measure for process cluster evaluation. The measure uses information derived by a static source analysis as well as information about interprocess communication during the application execution. Experiments show that the resulting quantitative evaluation conforms with a human evaluation of the same clusters