The composition of ad-hoc agent-oriented design processes

1
The composition of ad-hoc agent-oriented design
processes
Massimo Cossentino ICAR- Consiglio Nazionale
delle Ricerche Palermo Valeria
Seidita Dipartimento di Ingegneria Informatica
Univ. Palermo

• Miniscuola WOA08
• 18 Novembre 2008

2
Outline
3
Introduction
4
Outline
5
Introduction

• In this talk we will deal with a well known
  problem the construction of a design process
• When do we face this problem?
• While developing a new design process (several
  new processes are proposed in literature yearly)
• While improving an existing process
• Because it does not give good results (how do we
  measure that?)
• Because we want to apply it to a different class
  of problems, or in a different development
  context
• How do we solve the problem?
• Situational Method Engineering (SME) studies that
  from years
• We propose an extension of classical SME
  approaches that is specifically conceived for AOSE

6
What if my design process is not good?

• Each problem is suited for a specific class of
  problems
• For instance, PASSI has been conceived to develop
  information systems by using peer-agents
  societies with a classical (i.e. long but well
  documented) design process
• What if I am facing a different class of
  problems?
• I can use a different design process
• C. Bernon, M. Cossentino, J. Pavón. An Overview
  of Current Trends in European AOSE Research.
  Informatica Journal, vol. 29, Number 4, 2005
• L. Cernuzzi, M. Cossentino, F. Zambonelli.
  Process Models for Agent-based Development.
  Journal of Engineering Applications of Artificial
  Intelligence, (EAAI). Elsevier. Vol. 18, No.2,
  March 2005.
• I can create a new methodology for my specific
  purposes
• Situational Method Engineering
• M. Saeki. Software speci?cation design methods
  and method engineering. International Journal of
  Software Engineering and Knowledge Engineering,
  1994.
• B. Henderson-Sellers. Method engineering Theory
  and practice. In D. Karagiannis and editors Mayr,
  H. C., editors, Information Systems Technology
  and its Applications., pages 1323, 2006.

7
(Situational) Method Engineering
8
Outline
9
Method Engineering how it works

• The development methodology is built by the
  developer by assembling pieces of the process
  (method fragments) from a method base.
• The method base is composed of contributions
  coming from existing methodologies and other
  novel and specifically conceived fragments
• This is the approach used within the FIPA
  Technical Committee Methodology (2003-2005)

10
The normal agent development process
11
Situational Method Engineering
12
Adopting Method Engineering

• What do I need?
• A collection of method fragments
• Some guidelines about how to assemble fragments
• A CAME (Computer Aided Method Engineering) tool
• an evaluation framework (is my new methodology
  really good?)

13
The new process production
CAPE tool
14
SME approaches

• Brinkkemper S. (1996) Method Engineering
  Engineering of Information Systems Development
  Methods and Tools. Inf. Software Technol., 38(4),
  275-280.
• Brinkkemper S., Saeki M. and Harmsen F. (1998)
  Assembly techniques for method engineering,
  CAiSE'98, Proceedings, Springer Verlag, LNCS 1413
  pp. 381-400.
• Weerd I. van de, Brinkkemper S., Souer J.,
  Versendaal J. (2006) A Situational Implementation
  Method for Web-based Content Management
  System-applications Method Engineering and
  Validation in Practice. In Software Process
  Improvement and Practice (in press), John Wiley
  Sons, Ltd.
• Ralyté J. and Rolland C. (2001b) An approach for
  method re-engineering, Proceedings Int. Conf.
  ER2001, LNCS 2224, Springer-Verlag, Berlin,
  471-484
• Ralyté J. (2002). Requirements Definition for the
  Situational Method Engineering. Proceedings of
  the IFIP WG8.1 Working Conference on Engineering
  Information Systems in the Internet Context
  (EISIC02), Kluwer Academic Publishers,
  pp.127-152. Ralyté J. and Rolland C. (2001) An
  Assembly Process Model for Method Engineering.
  CAISE01, Proceedings, LNCS 2068,
  Springer-Verlag, pp. 267-283.
• Mirbel I. and Ralyté J. (2006) Situational Method
  Engineering Combining Assembly Based and
  Roadmap-Driven Approaches. Requirements
  Engineering, 11(1), pp. 5878.
• Nguyen V.P. and Henderson-Sellers B. (2003)
  Towards automated support for method engineering
  with the OPEN Process Framework. Procs. 7th
  IASTED Int. Conf. on Software Engineering and
  Applications, ACTA Press, Anaheim, CA, USA,
  691-696.
• Firesmith D.G. and Henderson-Sellers B. (2002)
  The OPEN Process Framework. An Introduction,
  Addison-Wesley, 330pp.
• Gonzalez-Perez C. (2005) Tools for an extended
  object modelling environment. In 10th IEEE
  International Conference on Engineering of
  Complex Computer Systems (ICECCS). IEEE Computer
  Society. 20-23.

15
Brinkkemper et als method fragment

• Method fragment is a coherent piece of
  information system development
• Two kinds of method fragment
• Process fragment
• Describes the stage, activities and tasks
• Product fragment
• Concerns the structure of a process product
  (deliverables, diagrams, etc.)
• Three orthogonal dimension
• Perspective
• The process and product perspective on fragment
• Abstract level
• Conceptual, technical and external level
• Layer of granularity
• The level of decomposition at which a method
  fragment resides
• Method, stage, model diagram and concept
• A fragment can be composed of other fragments and
  can have relationships with other fragments

16
Method Chunk - Ralytè et al.

• Method chunk (seen as a consistent and autonomous
  component)
• It represents a portion of process with its
  resulting work products
• It integrates the product and the process aspects
  of method fragment
• It is represented using a metamodel (UML notation)

17
Method Chunk Metamodel
18
Method Chunk Metamodel

• It is composed of two parts
• The process model
• Each chunk can be atomic (simple) or can
  aggregate other chunks
• Guideline, embodies the method chunk knowledge to
  guide the designer
• Situation, the condition of chunk applicability
• Intention, the objective to perform
• Descriptor, describes the situation in which the
  chunk can be reused
• The product model
• It is composed of Product Model, Product Part
  and Guideline
• For each method there is at least one product
• Guideline is also part of the product model, it
  describes how to generate a product

19
OPF method fragment

• It is part of existing methodologies and used to
  construct new ones
• It is generated and stored in a repository with
  all its guidelines basing on OPF Metamodel
• The Metamodel is composed of five main
  metaclasses
• Each metaclass produces a method fragment

20
OPF Metamodel
21
Other References on SME

• Kumar K. and R.J. Welke (1992) Methodology
  Engineering a Proposal for Situation-Specific
  Methodology Construction, in Challenges and
  Strategies for Research in Systems Development,
  W.W. Cotterman and J.A. Senn (eds.). John Wiley
  Sons Chichester, UK. p. 257-269.
• Rolland C., Prakash N. and Benjamen A. (1999) A
  multi-model view of process modelling,
  Requirements Engineering J., 4(4), 169-187.
• Saeki M. (2003) Embedding Metrics into
  Information Systems Development Methods An
  Application of Method Engineering Technique.
  CAiSE03, Proceedings, LNCS 2681, Springer,
  374-389.
• Saeki M. (2003) CAME the First Step to Automated
  Software Engineering. Procs. OOPSLA 2003 Workshop
  on Process Engineering for Object-Oriented and
  Component-Based Development. Anaheim, CA, 26-30,
  COTAR, Sydney, 7-18.
• Gupta D. and Prakash N. (2001) Engineering
  Methods from Method Requirements Specifications.
  Requirements Engineering Journal. 6(3), 135-160.
• Henderson-Sellers B. (2002) Process metamodelling
  and process construction Examples using the OPEN
  Process Framework (OPF). Annals Software Engin.
  14, 341362.
• Henderson-Sellers B. (2005) Creating a
  comprehensive agent-oriented methodology using
  method engineering and the OPEN metamodel,
  Chapter 13 in Agent-Oriented Methodologies (eds.
  B. Henderson-Sellers and P. Giorgini), Idea
  Group, 368-397.
• Henderson-Sellers B. (2006) Method engineering
  theory and practice, Information Systems
  Technology and its Applications. 5th
  International Conference ISTA 2006. Klagenfurt,
  Austria, LNI Proceedings, Volume P-84, Bonn,
  13-23.

22
Agent-Oriented Situational Method Engineering
23
MAS Meta-model

• MAS meta-models play a fundamental role in the
  design of agents.
• Different design processes have great differences
  in the MAS meta-models they adopt
• A meta-model is a model of the concepts that can
  be used to design and describe actual systems.
• Models describing a system are composed of
  elements that are instances of meta-model
  elements
• MAS meta-models usually include elements like
  role, goal, task, plan, communication,

24
Software Design the role of system meta-model

• Designing (a software) means instantiating its
  meta-model

META-MODEL
MODEL
25
The Prode (PROcess DEsign for design processes)
approach for Agent-Oriented Method Engineering
MMM
26
Method Fragment structure (result of the FIPA
Methodology TC work)
M. Cossentino, S. Gaglio, A. Garro, V. Seidita.
Method Fragments for agent design methodologies
from standardization to research. International
Journal on Agent Oriented Software Engineering
(IJAOSE). 1(1). 2007.
27
FIPA method fragment

• A fragment is a portion of the development
  process, composed by
• A portion of process (what is to be done, in what
  order), defined with a SPEM diagram
• One or more deliverables (artifacts like
  (A)UML/UML diagrams, text documents and so on).
• Some preconditions (they are a kind of constraint
  because it is not possible to start the process
  specified in the fragment without the required
  input data or without verifying the required
  guard condition)
• A list of concepts (related to the MAS
  meta-model) to be defined (designed) or refined
  during the specified process fragment.
• Guideline(s) that illustrates how to apply the
  fragment and best practices related to that
• A glossary of terms used in the fragment (in
  order to avoid misunderstandings if the fragment
  is reused in a context that is different from the
  original one)
• Other information (composition guidelines,
  platform to be used, application area and
  dependency relationships useful to assemble
  fragments)

M. Cossentino, S. Gaglio, A. Garro, V. Seidita.
Method Fragments for agent design methodologies
from standardization to research. International
Journal on Agent Oriented Software Engineering
(IJAOSE). 1(1). 2007.
28
The Process Design for Design ProcessPRODE
29
Outline
30
PRODE divided inthree main areas of research
MMM
31
PRODE Area 1A Collection of Process Fragments
32
The fragment collection in PRODE
MMM
33
The PRODE Process Representation
34
Applying the Proposed Method Fragment Definition

• A method Fragment can be explored from four
  points of view
• Process
• The process related aspect of the fragment
  workflow, activity and work product
• Storing
• It concerns with the storage of the fragment in
  the method base and its retrieval
• Reuse
• It concerns with the reuse feature of the
  fragment and lists the elements helpful in
  reusing the fragment during the composition of a
  new design process
• Implementation
• The implementation of the main elements of the
  process view.
• Method fragment construction is Work Product
  oriented, a method fragment must deliver a
  product.
• Cossentino, M., Gaglio, S., Garro, A. and
  Seidita, V. (2007). Agents and Method
  Engineering a Standardization Perspective, Int.
  J. Agent-Oriented Software Engineering, Vol. 1,
  No. 1, pp.91121.

35
Prode Area 2Guidelines for Fragment Assembly
36
Fragments Assembling in PRODE
MMM
2) Guidelines for fragment assembling
37
Process Analysis and Design in PRODE
38
Process Requirements Analysis
39
Process Analysis and Design in PRODE
40
Core Metamodel Creation
41
The ASPECS Core MetaModel
ASPECS is a design process for building holonic
multi-agent systems recently developed at UTBM
Massimo Cossentino, Nicolas Gaud, Stephane
Galland, Vincent Hilaire, and Abderrafiaa Koukam.
A Holonic Metamodel for Agent-Oriented Analysis
and Design. 3rd International Conference on
Industrial Applications of Holonic and
Multi-Agent Systems (HoloMAS'07). September 3 -
5, 2007, Regensburg, Germany.
42
Process Analysis and Design in PRODE
43
What is Prioritization ??

• The problem we face is
• What are the first fragments we should introduce
  in the new process?

??
44
The Algorithm

• Main issues
• We assume each process fragment instantiates,
  relates or quotes MAS MetaModel Elements (MMMEs)
• We created an algorithm for assigning a priority
  to the realization of some MMMEs
• Elements that are leaves of the metamodel graph
  are realised as first
• Other elements follow according to the number of
  their relationships
• The output is a priority list of fragments

45
The Prioritization Algorithm (1 of 3)

• Select a metamodel domain (consider the
  resulting metamodel as a graph with nodes (MMMEs)
  and edges (relationships))
• Define List elements1 as a list of MMMEs that can
  be defined by reusing fragments from the
  repository, and the associated priority p List
  elements1 (MMME, p), p1
• Define List elements2 as a list of MMMEs that
  cannot be defined by reusing fragments from the
  repository
• Define List elements3 as a list of elements that
  are not in the core MMM
• While the core MMM is not empty
• Select the leaves Li (i1,. . . ,n) that (i) can
  be instantiated by fragments of the repository
  and (ii) have less relationships with other
  elements
• Insert Li (i1,. . . ,n) in List elements1
• Remove elements Li (i1,. . . ,n) from the core
  MMM
• p p1
• While the core MMM is not empty
• Select the leaves Li (i1,. . . ,m) that can not
  be instantiated by fragments of the repository
• Insert Li (i1,. . . ,m) in List elements2
• Remove Li (i1,. . . ,m) from the core MMM

46
The Prioritization Algorithm (2 of 3)

• For each element E1i of List_elements1 select an
  instantiating fragment from the repository
  (verify the correspondence among fragment
  rationale and the process requirements/strategies)
  
• If one fragment corresponds to process
  requirements and strategies then
• insert the fragment in the new process
  composition diagram
• analyze inputs Ii (i0,. . . ,n) and outputs Oj
  (j0,. . . ,m) of the fragment
• If some Ii or Oj does not belong to the core MMM
  then add it to List_elements3 mark the fragment
  as To be modified
• remove E1i from List elements1
• For each element E2i in List_elements2 analyze if
  there is a similarity with the elements defined
  in this fragment
• if yes delete E2i from List_elements2 and Ii/Oi
  from List_elements3
• else (if no fragment correspond to requirements
  and strategies) then
• remove E1i from List_elements1 and insert it in
  List_elements2

47
The Prioritization Algorithm (3 of 3)

• For each E2i (i0..m) in List_elements2
• Define a new fragment for instantiating E2i
• Insert the fragment in the new process
  composition diagram
• Remove E2i from List_elements2
• For each E3i (i0..m) in List_elements3
• Introduce elements E3i (i0..q) from
  List_elements3 in the core MMM
• Repeat from 2. (consider only the new elements)
• If the process is not completed (i.e. not all
  design activities from requirements elicitation
  to coding, testing and deployment have been
  defined)
• Repeat from 1.

48
Process Analysis and Design in PRODE
49
The first two fragments in Building the ASPECS
Process
Not in the core metamodel

50
Process Analysis and Design in PRODE
51
The Process Component Diagram
52
Process Analysis and Design in PRODE
53
Metamodel Extension

• The Core MAS Metamodel is the starting point for
  selecting the right fragments from the repository
  and for assembling them in the new process
• MAS Metamodel extensions come from
• The need of incorporating MMMEs referred in
  selected fragments
• New process requirements
• Not all design activities from requirements
  elicitation to coding, testing and deployment
  have been defined
• Three different situations may arise
• Different MAS metamodels contribute to the new
  one with parts that are totally disjointed
• Different MAS metamodels contribute to the new
  one with parts that overlap and
• overlapping elements have the same definitions
  bounded to elements with different namesor on
  the contrary
• ...overlapping elements have the same name but
  different definitions

54
Method Engineering Supporting Tools
55
Method Engineering Supporting Tools

• MetaEdit1 is probably the most diffused tool
• It is a CAME (Computer Aided Method Engineering)
  tool. It instantiates a CASE (Computer Aided
  Software Engineering) tool.
• It offers no specific support for proper process
  life-cycle adoption, distributed work,
  collaboration
• Metameth is a recent research tool developed in
  Palermo
• It is a CAPE tool that instantiates a CASE tool.
• It is based on workflow concepts and supports
  collaborative distributed work. Designer is
  supported by an expert system.
• Eclipse Process Framework (EPF)3 is an open
  source effort
• It is a process modelling tool that adopts SPEM
  (Software Process Engineering Metamodel) by OMG.
• Not properly a method engineering tool but very
  useful in documenting/representing the process
• The Rational Method Composer tool is another
  example of this kind of tool
• 1 http//www.metacase.com/mep/
• 2 M. Cossentino, L. Sabatucci, V. Seidita, S.
  Gaglio. An Agent Oriented Tool for New Design
  Processes. Fourth European Workshop on
  Multi-Agent Systems (EUMAS'06). Lisbon, Portugal.
  December 2006.
• 3 http//www.eclipse.org/epf/

56
MetaEdit by MetaCASE

• MetaEdit is composed of two main components
• The Workbench for defining the modeling language
  and related diagrams. It uses the GOPPRR
  metamodeling language. This is the CAME part of
  MetaEdit
• The Modeler for actually designing the system.
  This is the CASE part of MetaEdit and allows the
  instantiation of the concepts and rules defined
  with the first step.

• Comments
• GOPPRR is powerful but defining a new process is
  a very demanding operation
• Expansion capabilities (for instance with
  plug-ins or new behaviour modules) are quite
  limited
• It is quite hard to convert it in a CAPE tool
  (see experience done with Agile PASSI)

57
PRODE Area 3Supporting Tools

• Metameth

58
PRODE divided inthree main areas of research
MMM
59
What is metameth

• Metameth is an (open-source) agent-oriented tool
  we built to support our experiments in
  methodologies composition and their application
  in real projects.
• Metameth is
• a CAPE tool since it supports the definition of
  the design process life-cycle and the positioning
  of the different method fragments in the intended
  place
• a CAME tool since it allows the definition of
  different method fragments
• a CASE tool since it supports a distributed
  design process, it offers several (by now UML)
  graphical editors and an expert system for
  verifying the resulting system

60
The Metameth tool architecture
61
Results Evaluation
62
Results Evaluation an open problem?
MMM
63
AO Design Process Evaluation

• Q.N. Tran, G. C. Low (2005). Comparison of Ten
  Agent-Oriented Methodologies. In Agent-Oriented
  Methodologies, chapter XII, pp. 341367. Idea
  Group.
• L. Cernuzzi, G. Rossi (2002). On the evaluation
  of agent oriented methodologies. In Proc. of the
  OOPSLA 2002 Workshop on Agent-Oriented
  Methodologies, pp. 21-30.
• Arnon Sturm, Dov Dori, Onn Shehory (2004). A
  Comparative Evaluation of Agent-Oriented
  Methodologies, in Methodologies and Software
  Engineering for Agent Systems, Federico Bergenti,
  Marie-Pierre Gleizes, Franco Zambonelli (eds.)
• Khanh Hoa Dam, Michael Winikoff (2003). Comparing
  Agent-Oriented Methodologies. In proc. of the
  Agent-Oriented Information Systems Workshop at
  AAMAS03. Melbourne (AUS).
• P. Cuesta, A. Gómez, J. C. González, and F. J.
  Rodríguez (2003). A Framework for Evaluation of
  Agent Oriented Methodologies. CAEPIA'2003
• L. Cernuzzi, M. Cossentino, F. Zambonelli (2005).
  Process Models for Agent-Based Development.
  International Journal on Engineering Applications
  of Artificial Intelligence (EAAI). Elsevier.

64
Details on AO processes evaluation

• From
• Q.N. Tran, G. C. Low (2005). Comparison of Ten
  Agent-Oriented Methodologies. In Agent-Oriented
  Methodologies, chapter XII, pp. 341367. Idea
  Group.

Structure of the evaluation framework
65
Details on AO processes evaluation/2

• From
• Arnon Sturm, Dov Dori, Onn Shehory. A Comparative
  Evaluation of Agent-Oriented Methodologies, in
  Methodologies and Software Engineering for Agent
  Systems, Federico Bergenti, Marie-Pierre Gleizes,
  Franco Zambonelli (eds.)
• Evaluation is based on
• concepts and properties (autonomy,
  proactiveness,, ),
• notations and modeling techniques (accessibility,
  expressiveness),
• process (development context, Lifecycle
  coverage),
• pragmatics (required expertise, scalability, ) .

66
Details on AO processes evaluation/3

• From
• Khanh Hoa Dam, Michael Winikoff (2003). Comparing
  Agent-Oriented Methodologies. In proc. of the
  Agent-Oriented Information Systems Workshop at
  AAMAS03. Melbourne (AUS).
• Based on a questionnaire
• Reused and extended in AL3-AOSE TFG3(see
  website1 forresults)

1 http//www.pa.icar.cnr.it/cossentino/al3tf3/cont
ributions.html
67
Details on AO process evaluation/4

• The Capability Maturity Model Integration (CMMI)
• The overall goal of CMMI is to provide a
  framework that can share consistent process
  improvement best practices and approaches, but
  can be flexible enough to address the rapidly
  changing needs of the community.
• SCAMPI (Standard CMMI Assessment Method for
  Process Improvement) it is a schema for process
  evaluation in five steps activation, diagnosis,
  definition, action, learning.

68
Details on AO process evaluation/5CMMI discrete
levels

• Levels are used in CMMI to describe an
  evolutionary path recommended for an organization
  that wants to improve the processes
• The maturity level of an organization provides a
  way to predict an organizations performance in a
  given discipline or set of disciplines.
• A maturity level is a defined evolutionary
  plateau for organizational process improvement.

69
Details on AO process evaluation/6CMMI discrete
levels
Maturity Level Description
1-Initial processes are usually ad hoc and chaotic
2-Managed processes are planned and executed in accordance with policy
3-Defined processes are well characterized and understood, and are described in standards, procedures, tools, and methods
4-Quantitatively managed the organization and projects establish quantitative objectives for quality and process performance and use them as criteria in managing processes
5-Optimizing an organization continually improves its processes based on a quantitative understanding of the common causes of variation inherent in processes

• AOSE processes are (at most) at level 3!!
• (only a few of them)

70
Open issues

• SME is perceived to be a difficult discipline
• This is only partially true. All new design
  processes creator performed (usually in a
  disordered way) the steps proposed and studied by
  SME
• A greater diffusion of AO-SME can have positive
  effects on the development of new AO design
  processes (specifically in new areas like
  self-org)
• Major problems with AO-SME
• AO processes deals with MAS metamodels and they
  are an open issue in the agent community
• Lack of standards (ISO specification vs FIPA
  proposal)
• Lack of standard repository of fragments
• Lack of stable (commercial quality) CAPE/CAME
  tools
• Design process evaluation is still an open issue
  in both AO and OO software engineering.

71
Fragment RepositoryMetameth Tool

• V. Seidita

72
Appendix

• More slides on selected topics

73
The PASSI Process Life-Cycle
74
PASSI Models Scope

• Each Model addresses a specific concern
• System Requirements Model
• It aims at defining system functionalities and
  assigning them to agents
• Agent Society Model
• It aims at defining agent social features like
  communications, domain knowledge, agent roles.
• Agent Implementation Model
• It defines the agent solution according to the
  selected implementation platform
• Code Model
• It includes code obtained by pattern reuse and
  manually created code
• Deployment Configuration Model
• It defines dependencies among agents and host
  configurations in multi-hosts/mobile agents
  applications

75
PASSI ToolKit (PTK) screenshoots
76
PASSI new modelling notation
77
The New Unified Notation
78
Example of new notation Task Specification
Diagram
79
Example of new notation Communication Ontology
Description diagram
80
Example of new notation Role Description Diagram
81
Classical vs. agent-oriented method engineering
(PRODE approach)

• Other authors work on the field (B.
  Henderson-Sellers, J. Ralyte, S Brinkkemper) but
  PRODE is specifically agent-oriented and focused
  on the importance of the MAS metamodel
• A comparison of those approaches with mine can be
  found in
• Seidita, V., Ralyte, J., Henderson-Sellers, B.,
  Cossentino, M., Arni-Bloch, N. A comparison of
  deontic matrices, maps and activity diagrams for
  the construction of situational methods. The 19th
  International Conference on Advanced Information
  Systems Engineering (CAiSE'07) Forum. 11-15 June
  2007, Trondheim, Norway.