The FEAST/1 project (EPSRC grant GR/K86008) began its investigations in October 1996 and terminated at the end of September 1998. Its broad objectives were to construct black and white box models of software system evolution, with special attention to feedback phenomena. To ensure practical relevance, such models need to reflect actual projects. They must, therefore, be derived from metric data and expertise accumulated in industrial evolution processes. Such data has indeed been provided by the FEAST collaborators¹. Their active interest and participation has been crucial to the major progress made.

Further data sources have also emerged. Through the good offices of Dr D E Perry², Bell Labs - Lucent Technologies has provided important data on two of its switching systems. In addition BT has expressed strong interest in the work and nominated a candidate system for study in FEAST/2. They are providing a Short Term Research Fellowship to one of the FEAST/1 RAs to bridge the gap between FEAST/1 and /2.

A significant part of the FEAST/1 goals have already been achieved and documented (sects. G, J). Ever more evidence of the impact of feedback on the software process is accumulating. The results indicate that there is sufficient discipline, structure and dynamics in software evolution to warrant systematic approaches to study of the phenomenon. The data analysis and modelling has revealed common patterns of evolution across those of the systems studied [leh98e] that are release based evolving systems and not ab initio developments. Their long term growth trends, for example, are closely modelled by inverse square relationships [tur96]. Growth patterns have been observed that are similar to those previously reported [leh80,85b]. Moreover, these observations are largely consistent with the laws of software evolution as originally formulated [leh74 - 85b]. During the development of various system dynamics (SD) models, influential feedback paths have been identified [leh98c]. One such model faithfully reproducing long term behaviour of ASRAAM evolution has recently been described [wer98]. With intensive participation by ICL personnel an SD model of the VME Kernel evolution process is being calibrated [leh98c]. Overall the results are already impacting collaborator practices within and outside the direct contact groups. Confidence in the FEAST premises, both among the industrial collaborators and the researchers, is now strong.

Current studies of software evolution must be extended to further advance understanding and mastery of the phenomenon. FEAST/2 will focus on control and exploitation of process behaviour. It will determine more fully the implications of the FEAST hypothesis and its significance for the management of software evolution and its support. To this end the project will refine existing models, drawing on new observations and by refining models to a level at which the impact of management controls becomes explicit. FEAST/2 will also establish technical criteria for management decision support, planning and tools to support software evolution and process improvement. This will open up the way for industry wide exploitation. However, the collaborators have suggested that we concentrate on research issues and exclude development of tools of this type from the FEAST/2 program. They see the development of such tools as requiring strong familiarity with organisational concerns and culture as well as with methods, procedures and tools being used elsewhere in the organisation. Given technical guidelines, it is therefore best undertaken internally.

Progress made in understanding the impact of feedback in the global software process has inevitably led to recognition of new issues and others requiring further clarification. On the basis of what has been learned from FEAST/1, the following represent those of most immediate theoretical interest and practical concern.

FEAST/2 cannot reasonably expect resources to undertake all these candidate tasks. The list has, therefore, been pruned to omit those which, if not addressed still permit the project objectives to be attained and the remaining tasks to be fully investigated. It is proposed to concentrate on tasks i - iv, vi - ix, and xi. Task v is omitted as per the suggestion of the current collaborators. Task x is excluded because more understanding of the influences and mechanisms that determine evolutionary behaviour must be achieved before one can hope to develop generic models. Though they reflect issues of significant concern and opportunity, tasks xii to xv cannot be undertaken in FEAST/2 because of the additional resources their inclusion would require.

The selected objectives underlying the subset of FEAST/2 tasks identified above are as follows

In addition to further study of research issues already studied, the following will be addressed in FEAST/2:

The topic and exploitation of Process Modelling [ost87, fin95, der98] and Software Metrics [kit96, fen97] have been of interest to sections of the software engineering community for many years. Empirical Studies are of more recent vintage [har97]. The FEAST focus on software evolution and feedback, local and global, is, however, believed to be innovative and unique, posing its own particular problems. We shall therefore continue to monitor work in related areas and evaluate approaches demonstrating a potential for advancing our investigations, refining as necessary and seeking to exploit those that show promise.

Black and white box modelling, have proven fundamental to and fruitful in our investigations to date. The former provides means for revealing evolutionary behaviour. The latter seeks to identify the mechanisms that cause and control it, to determine their attributes and to replicate that behaviour. Extensive use has also made of statistical analysis techniques [man91]. Other approaches being considered are described below.

FEAST/1 black box models have shown the influence of global dynamics on software processes. FEAST/2 will continue to focus on externally visible behaviour but also include decision making processes. Its key ingredients will be measurement, visualisation, analysis, model building and interpretation. Visualisation of measures and attributes, including time, augmented by appropriate statistical analyses [man91], provide insights into system evolution as depicted by the historical data of the systems studied. British Telecom have suggested inclusion of economic indicators and financial data in the analysis of their candidate system. The combined techniques provide a basis for model development and identification of patterns and trends. Their interpretation will be in terms process phenomenology.

The white box investigation focuses on the internal process structure, seeking to identify feedback mechanisms and their properties. Its key ingredients are interpretation, modelling, calibration and validation, i.e. establishing accuracy of model behaviour and that of its internal mechanisms relative to the real world.

Most current software process modelling is enactment oriented [fin94, der98]. It highlights individual activities and relationships between them using using approaches such as graphical notations [oul95], rule based notations [bar92] and programming-language-like formalisms [ost87]. Such activity-focussed techniques have many uses, but are not ideal for investigating feedback effects, effects largely ignored by the mainstream process modelling community. To investigate them, systems dynamics [ven95] modelling is more appropriate. The FEAST/1 top down approach [zur69] to modelling global processes [leh98c,wer98] and its extension in FEAST/2 by successive refinement is, to the best of our knowledge, wholly innovative.

The collaborators have recognised that FEAST/1 results constitute a preliminary source of management rules, guidelines and limits. They regard these and the potential for more to come as a major contribution to software process improvement. At their request, a first decision support set based on current results will be prepared as a text document and delivered in the autumn. In view of its wider significance and the research issues raised, the topic is being explicitly included in the FEAST/2 objectives (sect. C.2 (a)). This requires introduction of machine support into the extraction, analysis, interpretation, presentation and dissemination of the rules etc. developed. Extraction and analysis will be supported by the approaches outlined in the next subsection. As proposed in section B ii, the other steps will, in collaboration with Professor M Sloman, be facilitated by the use of a notation [lup97] developed in the Department under EPSRC contract GR/K37512. The work will serve as a demonstrator for that technology. It is of interest to note that adoption of the decision support [pin96, rus98] findings by the collaborators changes their processes. FEAST/2 will have become a feedback element in their processes that may have to be represented in the models.

Until recently only very small (12 - 28 point) data sets representing aggregated global behaviour were available to FEAST/1. Very much larger data sets, from configuration management systems for example, have now become available. This suggests that a machine-assisted approach to data association, model building and behavioural analysis is necessary. Dr Y Guo, Technical Director of the Department's Parallel Computing Centre and Principal Investigator of a data mining project funded by Fujitsu has suggested that data mining techniques [cha97] would help cleansing and analysis of these historical databases and assist in obtaining greater insight into the evolution of the systems being studied. Its use in FEAST/2 will add method and machine support to data pattern recognition and so facilitate the pursuit of task B i.

FEAST/2 objectives include investigation of the impact of feedback and control by people in both technical and non technical, global and local roles (section B xi). Such human involvement [leh98d] and other aspects of the problems [lev90, fat98] being studied suggest the use of fuzzy dynamics modelling [fri96,97]. Profs. Friedman and Sandler of the Jerusalem College of Technology, internationally recognised authorities on fuzzy dynamics and its use in modelling system evolution, have expressed strong interest in collaborating in this investigation. Their involvement in FEAST/2 will assist the processes of analysis, modelling and prediction. Contact with Professor EáH Mamdani [mam74] of IC EE will also be helpful. Fuller justification for the use of fuzzy dynamics in the context of FEAST/2 may be found in [ram98].

The central role of humans, individually and as teams, in all aspects of software evolution including all levels of management, marketing, usage and user support also suggests the use of multi-agent modelling techniques. The Department's April language [mcc95] will be used to build such models in the context of task B i, xi and related activities. Evaluation of multi agent techniques in the study of software evolution is included in a current PhD project involving Profs. KáLáClark and MáMáLehman and Dr B Nuseibeh.

FEAST/2 will build on a conceptual and theoretical base already established. Consolidation and extension of the latter requires four separate but closely related work packages. The attached Case for Support, Part 3 shows the composition of each at the next level of detail, the timings for each sub task and the timings, as indicated by diamonds, for milestone deliveries, generally reports, papers or other documentation.

The FEAST/1 collaborators have demonstrated their appreciation of the project's relevance to them by active participation in the three-monthly workshops and by their desire and commitment, as per attached letters of support, to participate in FEAST/2. Moreover, Matra-BAe, ICL and Logica have each requested on site presentations to a wider audience of their colleagues to spread FEAST/1 results in their organisations. Appreciating the benefits FEAST technology can bring, British Telecom have taken the initiative and requested involvement in FEAST/2 as has the DERA Software Engineering Centre. We look forward to the contributions these organisations will make to FEAST/2. Other approaches have also been made but could not be accommodated without requiring resources beyond those to be reasonably requested.

All this testifies to the industrial benefits that participation is seen as bringing. Such benefits include models and better understanding of their respective processes, the potential for process improvement resulting therefrom; the availability of formal management descriptions of policy elements and procedural rules; and of models for evolution planning, monitoring, prediction and process evaluation. The collaborators regard the development of tools to be derived from all these to be their responsibility since any such tools must be integrated readily with methods and tools already in use.

With general dissemination as outlined in G software development organisations in the UK and elsewhere will be able to adopt the FEAST techniques and so achieve similar benefits⁷. Realisation of this wider potential requires documentation of the means by which FEAST results have been obtained. Its availability will enable individual organisations to apply the technology to the improvement of their processes. Relevance of the study outside to business processes in general is a matter for future investigation.

17 journal and conference publications, 7 book chapters⁸, 15 conference/ workshop presentations, 5 keynote/ invited talks and 18 lectures to University and industry groups and the responses from these testify to the degree of dissemination achieved in FEAST/1. A similar level of activity, at least, is expected in FEAST/2.

FEAST/1 has attracted wide international interest. Related work is already underway at the U. of York and Glasgow Caledonian U in the UK and at various locations in the USA and Israel. An example of dissemination activity is the key role played in the recent ProSim (Software Process Simulation and Modelling) workshop. Professor Lehman gave a Keynote lecture and, jointly with Dr Wernick (RA) presented a second paper. As a result the latter has joined the organising/program committee for ProSim '99 and Imperial College is being considered as the location for that meeting.

The Principal Investigator will devote himself full time to the project. The two RAs, will spend some 15% of their time working with collaborator personnel on site to gather data, discuss findings and conclusions. While continuing their current close collaboration, one will focus on black the other on white box activity.

All co-investigators and other named individuals expect to be actively involved in the project. Professor B Rustem will advise on development of the decision support activity. Dr Nuseibeh will collaborate in the multi agent modelling of requirements analysis and system specification and incorporation of these processes into the refined SD models. Professor M Sloman will supervise development of formal rule and policy documentation. For Professors Friedman and Sandler two five day meeting each year, one in the UK, one in Israel, are planned. UK meetings will overlap FEAST workshops.

The collaborators will each contribute a significant number of person-days of effort, mainly technical, to the project at a level as indicated in their respective letters of support. They are also committed to full participation in the planned quarterly workshops.

Theoretical foundations for FEAST continue to be developed by a core group consisting of Professor W M Turski (University of Warsaw), Dr Dewayne Perry (Bell Labs - Lucent Technologies) and the Principal Investigator, the latter having day to day responsibility. The work is based on email contacts and periodic meetings supported by an EPSRC SVF grant, currently GR/L96561. The joint investigators, Professor. B Rustem and Dr V Stenning, will liaise closely with the Principal Investigator and participate in the work of the core group as appropriate.

Overall direction and day to day project management, including progress monitoring and provision of deliverables, will be the responsibility of the Principal Investigator supported by a part time secretary. The co-investigators will provide continuing support, for example in the critical appraisal of approaches being used, project plans, conclusions drawn and draft reports and papers.

Project workshops will be held at quarterly intervals, attended by the four investigators, project staff, the SVFs, representatives of the industrial collaborators and invited guests as appropriate. Each such meeting will review results to date to ensure that the project remains technically sound and plans for the remainder of the project to ensure that objectives are met. It will also enable collaborators to take away results to be explored for immediate relevance, tool building and/or other exploitation.

The black and white box activities, including continual interaction with collaborators, each require the full time attention of one person. Thus two RAs are required. On the basis of their contribution to and performance in FEAST/1 and recognising the value of their accumulated experience it is proposed to rehire the incumbents. They will be reappointed at one point above their present scale point where not already at the top of the appropriate scale. In view of the number of meetings and workshops to be organised, the level of publications, the number of visitors, the level of travel and the level of correspondence it is proposed to appoint a half time secretary.

Very close liaison with the industrial collaborators will continue as a central feature of FEAST/2. To achieve this, the two RAs will need to regularly travel to and work at collaborator sites. Funding is, therefore, requested for a total of 140 days (7 person days per month for 20 months) travel to collaborators' sites with some 20% involving overnight stays at a cost of ú40 and ú80 per day for one and two day trips respectively for a total of ú6720. Eight project workshops at IC will cost ú800.

The widespread and increasing interest in software evolution requires our participation in software engineering conferences such as ESEC, ICSE and ICSP and various process and evolution workshops. Such trips will also be used to visit other interested organisations in industry and government. Based on the activity in FEAST/1 as shown by the data in section G, support is requested for eight US trips (£9600), eight European/Israel trips (£6400) and ten UK trips (£1000). All costs include conference registration fees.

Three 21" colour screens are required to provide adequate displays for the modelling work. One additional desk top processor is required to replace outdated and inadequate equipment. Two ZIP drives are required for data interchange with collaborators. The consumables budget includes £2500 for new software (S-Plus, MATLAB, Data Engine mining tool, etc.) and upgrades, computer consumables and maintenance, systems support, connection to and use of the Department's network and the cost of document reproduction.

As indicated in the attached letters of support the industrial collaborators will provide significant staff effort to work on FEAST activities, travel and subsistence for these staff to attend meetings and workshops, and on-site working facilities for the FEAST investigators.

Copies of the more recent papers can be obtained from publication links at http://www.doc.ic.ac.uk/~mml/

[bar92] Barghouti NS and Kaiser GE, Scaling Up Rule Based Development Environments, Int. J. on Softw. Eng. and Knowledge Eng., v.2, n. 1, March 1992, pp. 59 - 78

[cha97] Chattratichat J, Guo Y et al, Large Scale Data Mining: Challenges and Responses, Proc. KDD'97, 3rd. Int. Conf. on Knowledge Discovery and Data Mining, Aug., 1997, Los Angeles, USA

[der98] Derniame J-C, Warboys B, and Kaba AB (eds.), Software Process,: Principles , Methodology, Technology, Manuscript in preparation

[fat98] Fatehi K, Modelling the Impact of Quality Improvement Initiatives, Proc. ICSE'98, v. 2, Apr. 19 - 25, 1998 Kyoto, Japan, pp. 164 - 165

[fen97] Fenton NE and Pfleeger SL, Software Metrics: A Rigorous and Practical Approach, Int. Thomson Comp. Press, London, 1997, 638 pps.

[fin94] Finkelstein A, Kramer J and Nuseibeh B, Software Process Modelling and Technology, Research Studio Press, Wiley , 1994, 384 pps.

[fri96] Friedman Y and Sandler U, Evolution of Systems Under Fuzzy Dynamic Laws, Fuzzy Sets and Systems, Vol. 84, 1996, pp. 61 - 74

[fri97] Friedman Y and Sandler U, Dynamics of Fuzzy Systems, inv. paper, to appear, Int. J. Chaos Theory and Application, 21 pps.

[har97] Harrison R (ed.), Proc. of ICSE'97 Workshop on Process Modelling and Empirical Studies of Software Evolution, May 18, 1997, Boston, USA, 95pps.

[kit96] Kitchenham B, Software Metrics - Measurement for Software Process Improvement, NCC Blackwell, 1996, 241 pps.

[leh74] Lehman MM, Programs, Cities, Students, Limits to Growth?, Inaugural Lecture, May 1974. Publ. in Imp. Col. of Sc. Tech. Inaugural Lect. Ser., vol 9, 1970, 1974, pp. 211 - 229. Also in Programming Methodology, (D Gries ed.), Springer, Verlag, 1978, pp. 42 - 62

[leh80] id., Programs, Life Cycles and Laws of Program Evolution, Proc IEEE Spec. Iss. on S.E., vol. 68, no. 9, pp. 1060, 1076. Sept 1980

[leh85b] Lehman MM and Belady LA, Program Evolution - Processes of Software Change, Acad. Press, London, 1985, pp. 538. The book includes much of the relevant material published up to the time of publication

[leh98c] Lehman MM and Wernick P, System Dynamics Models of Software Evolution, Processes, Proc. Int. Wrkshp. on the Princ. of Softw.Evolution IWPSE-98, ICSE-20, April 20-21, 1998, Kyoto, Japan, pp. 6-10

[leh98d] Lehman MM, Feedback, Evolution and Software Technology - The Human Dimension , ICSE 20 Workshop on Human Dimension in Successful Software Development, 20 Apr 98, Kyoto, Japan

[leh98e] Lehman MM and Ramil JF, The Impact of Feedback in the Global Software Process, Keynote Lect., ProSim '98 Workshop, Silver Falls, OR, 23 June 1998, to be publ. in J. of Sys. and Softw. 1999

[leh98f] Lehman MM and Ramil JF, Implications of Laws of Software Evolution on Continuing Successful Use of COTS Software, ICSTM, Dept. of Computing, Res. Report 98/8, Apr. 98

[leh98g] Lehman MM, Perry DE and Ramil JF, On Evidence Supporting the FEAST Hypothesis and the Laws of Software Evolution, to appear in Proc. Metrics'98, Bethesda, Maryland, Nov. 20 - 21, 1998

[lev90] Levary RR, System Dynamics with Fuzzy Logic, Int. J. Sys. Sci., 1990, v.21, n.8, pp. 1701-1707

[lup97] Lupu E and Sloman M, A Policy Based Role Object Model, EDOC'97, First Intl Enterprise Distr. Object Comp. Wrksh., Gold Coast, Queensland, Australia, Oct. 1997, pp 36-47.

[mcc95] McCabe FG and Clark KL, Programming in April: An Agent Process Interaction Language, Intelligent Agents, Springer Verlag, 1995.

[mam74] Mamdani EH, Application of Fuzzy Algorithms for Control of Simple Dynamic Plant, Proc. IEEE, v. 121, 1974, pp 1585-1588

[man91] Manly BFJ, Randomisation and Monte Carlo Methods in Biology, Chapman and Hall, London 1991, 281 pps.

[ost87] Osterweil L, Software Processes are Software Too, Proc. 9th Int. Conf. on Softw. Eng., Monterey, CA, 30 March - 2 Apr. 1987, IEEE Comp. Soc. Pub. n. 767, IEEE Cat. n. 87CH2432-3, pp. 2 - 13

[oul95] Ould MA, Business Processes - Modelling and Analysis for Re-engineering and Improvement, Wiley, Chichester, 1995

[pin96] Pinto R and Rustem B, Building and Solving Multicriteria Decision Models Involving Logical Conditions, in Computational Economic Systems - Models Methods & Econometrics, M. Gilli (ed.), Kluwer Acad. Pubs., Netherlands, 1996, pp. 123 - 136

[ram98] Ramil JF and Lehman MM, Fuzzy Dynamics in Software Project Simulation and Support, to appear, Proc. 6th European Worksh. on Softw. Process Technology, EWSPT 6, 16 - 18th Sept. 1998, Weybridge, Surrey, UK, LNCS 1487, Springer-Verlag, pp. 122-126

[rus98] Rustem B, Algorithms for Nonlinear Programming and Multiple Objective Decision, Wiley, 1998, 450 pps.

[tur96] Turski WM, A Reference Model for the Smooth Growth of Software Systems, IEEEáTrans. Softw. Eng., v. 22, n. 8, Aug. 1996, pp. 599 - 600

[ven95] Vensim Reference Manual, version 1.62, Ventana Sys. Inc. Belmont, MA, 1995

[wer98] Wernick P and Lehman MM, Software Process White Box Modelling for FEAST/1, ProSim '98 Workshop, Silver Falls, OR, 23 June 1998, to be publ. in J. of Sys. and Softw. 1999

[zur67] Zurcher FW and Randell B, Iterative Multi-Level Modelling - A Methodology for Computer System Design, Info. Proc. 67, Proc. IFIP Congr. 1968, Edinburgh, Aug. 1968, pp. D138 - 142

[¹] Data has been obtained from all four formal collaborators: ICL (VME Kernel), Logica (FastWire), Matra BAe, MoD DERA and also from Lucent Technologies.

[²] Dr Perry together with Professor WM Turski are associated with the FEAST project via the EPSRC SVF scheme under grants GR/L07437 (1996/7) and GR/L96561 (1998/9).

[³] For example by use of the notation [lup97] developed in the Department (IC DoC) with EPSRC and EC support

[⁴] Conceptual foundations for such procedures and tools exist [ram98].

[⁵] Short term fluctuations in long term trends have been shown to display a degree of structure.

[⁶] Requested by collaborators to permit them to extend use of models beyond the specific projects from which they were derived

[⁷] And so strengthen confidence in the theory and the practical techniques still further

[⁸] Including Springer LNCS