EPSRC Network on Evolvability in Biology & Software Systems

Software Evolution and Evolutionary Computation Symposium Abstracts

University of Hertfordshire, Hatfield, U.K.

7-8 February 2002

An Approach to the Prediction of Software Evolutionary Attributes and some of its Implications

J. F. RAMIL

Computing Dept., Faculty of Maths and Computing, The Open University, Walton Hall, Milton Keynes MK7 6AA, U.K.

j.f.ramil@open.ac.uk

M. M. LEHMAN

Department of Computing, Imperial College of Science, Technology and Medicine, 180 Queen's Gate, London SW7 2BZ, U.K.

mml@doc.ic.ac.uk

Software evolution, the process of software change, encompasses all activities performed over the entire lifecycle of a software system to maintain stakeholders' satisfaction within expectations. Such activities are, in a world increasingly dependent on computers and, therefore its software, of great social and economic importance. However, in the absence of sound methods, the long-term planning of evolution presents considerable challenges for managers in business and other organisations. After briefly enumerating some of the constraints under which software evolution (and its planning) occurs, the talk will introduce a sequence of steps to achieve mathematical models for resource estimation. The models relate effort applied to attributes such as product size and process work-rate. The steps are exemplified by a case study and will include numerical results on the predictive accuracy of the models. The modelling steps represent an instance of a wider iterative modelling framework that addresses intrinsic characteristics of the evolution process. Such a framework appears to be useful for monitoring and predicting the behaviour of open systems/processes that are characterised by a rich information/control feedback loop structure. Such systems/processes cannot, in general, be properly and effectively split into behavioural sub-systems. Moreover, they may exhibit patterns of intermittent behaviour, that is, periods of regularity separated by break-points or behavioural transitions. All this poses formidable challenges to accurately predict evolutionary behaviour. The modelling steps presented in this talk are believed to address some of the challenges. Moreover, they offer an approach to the wider problem of predicting the performance of other types of systems that involve human intelligence and action as essential constituents.

Homepages: http://mcs.open.ac.uk/jfr46,   http://www.doc.ic.ac.uk/~mml/