1. Loris PenseriniComputer Science Institute - University of Ancona 1 A distributed agent architecture and case-based approach for information system integration Computer Science Institute University of Ancona In collaboration with: Loris Penserini (pense@inform.unian.it) Maurizio Panti (panti@inform.unian.it) Luca Spalazzi (spalazzi@inform.unian.it) (these slides are available at: http://www.cs.toronto.edu/~mkolp/tropos/)

2. Loris PenseriniComputer Science Institute - University of Ancona 2 Motivation In a network environment the principal problems of Information System Integration are: - distribution: the needing of sharing sources between LANs and/or WANs - dynamism: new sources can be inserted or deleted, the schemas can be modified and so on - heterogeneity: discrepancies in physical, logical, and conceptual levels

3. Loris PenseriniComputer Science Institute - University of Ancona 3 Work Purpose - to use an Agent Platform to face the distribution - to solve dynamic problems by a Distributed CBR approach - to adopt a Mediator architecture to face the heterogeneity - to evaluate both the efficacy and the efficiency of the Mediator cooperation strategies by a prototype implementation

4. Loris PenseriniComputer Science Institute - University of Ancona 4 Summary Agent Management System Directory Facilitator Agent Communication Channel Agent Software * Agent Platform architecture FIPA compliant Mediator1 Case Memory1 * Mediator agent that uses a Distributed CBR Wrapper 1 db1 Wrapper 2 db2 * Mediator prototype and Cooperation Strategies Mediator2 Case Memory2

5. Loris PenseriniComputer Science Institute - University of Ancona 5 Agents AP 1AP 2 AP 3 The network Agent Platform (AP) architecture. DF AMS ACC IIOP FIPA Reference Model ? Agents ACC AMS DF ? Agents DF AMS ACC ? JEAP&FIPA

6. Loris PenseriniComputer Science Institute - University of Ancona 6 Communication Protocol FIPA defines two types of communication: Inter-platform Intra-platform Internet Inter-ORB Protocol No constraints CORBA allows several mechanisms in order to locate distribute objects, which are incompatible each other, but FIPA does not specify which one must be used. FIPA Specification: JEAP&FIPA

7. Loris PenseriniComputer Science Institute - University of Ancona 7 Communication Protocol AP prototype implementation: Intra-platform protocol: IIOP IIOP is a well known standard. Code Portability. ACC is simpler. JEAP&FIPA

8. Loris PenseriniComputer Science Institute - University of Ancona 8 Communication Protocol AP prototype implementation: CORBA mechanisms Wide compatibility with other FIPA compliant platforms. Interoperable Object Reference (IOR) Transient Name Service (TNS) JEAP&FIPA

9. Loris PenseriniComputer Science Institute - University of Ancona 9 ACC and AMS strictly interact each other. ACC supports two kinds of request: implicit ACC request; explicit ACC request (forward action); Agent Communication Channel (ACC) and Agent Management System (AMS) Agent Communication Channel (ACC) and Agent Management System (AMS) FIPA Specification: JEAP&FIPA

10. Loris PenseriniComputer Science Institute - University of Ancona 10 ACC and AMS (request :sender agent_i@iiop://192.168.1.1:800/acc :receiver acc@iiop://192.168.1.1:800/acc :language sl0 :ontology fipa-agent-management :protocol fipa-request :content (action acc@iiop://192.168.1.1:800/acc (forward (MESSAGE) )) :reply-with 0) A second source of overload is the task that ACC must do for explicit requests: Example: JEAP&FIPA

11. Loris PenseriniComputer Science Institute - University of Ancona 11 ACC and AMS 10: authenticate 8: request 3: agree 2: request 5: inform 4: authenticate 7: agree for Agent_i Agent_jACCAMSAgent_i 1: request for Agent_j 6: request from Agent_i 9: agree 11: inform 13: inform done for Agent_i 16: authenticate 14: request 15: agree 17: inform 18: inform done from Agent_j 12: agree from Agent_j JEAP&FIPA

12. Loris PenseriniComputer Science Institute - University of Ancona 12 It removes the interaction between ACC and AMS. AP prototype implementation: ACC and AMS are realised as a single agent: It does not decrease fault-tolerance capability of the platform. It is still FIPA compliant. ACC and AMS JEAP&FIPA

13. Loris PenseriniComputer Science Institute - University of Ancona 13 We only use implicit requests. ACC and AMS AP prototype implementation: It reduces the overload for the ACC. It is still FIPA compliant. JEAP&FIPA

14. Loris PenseriniComputer Science Institute - University of Ancona 14 Directory Facilitator (DF) The DF provides a yellow-pages service. The agents registered at the DF are classified by means of service- type and service-name. A Domain is the set of all the agents registered at the DF. The Agent Universe is the union of all the domains. Each Agent Platform must have at least one DF (the default DF). FIPA Specification: JEAP&FIPA

15. Loris PenseriniComputer Science Institute - University of Ancona 15 FIPA Platforms and Domains JEAP&FIPA

16. Loris PenseriniComputer Science Institute - University of Ancona 16 Directory Facilitator (DF) AP prototype implementation: Specification are not clear on what default DF can contain and which kind of service-name it must use. Logic structure of the default DF. Agent DF Level (GUID) Service Type Level Service Name Level default DF FIPA–ACC FIPA-DF FIPA – AMS Railway DF_j DF_ k … Airport DF_i JEAP&FIPA

17. Loris PenseriniComputer Science Institute - University of Ancona 17 Directory Facilitator (DF) AP prototype implementation: example A possible situation in which Agent_i would like to use a service that can perform the Wrapper_i, but at the begin Agent_i doesn’t know that. 6:Results 5:Invoke the service 3:Search for an agent 1:Search for a domain Agent_k Agent_i Agent Platform DF Defaul t DF_j 2:Answer with DF_j address 4:Answer with agent address (Agent_k) JEAP&FIPA

18. Loris PenseriniComputer Science Institute - University of Ancona 18 Some Remarks on FIPA FIPA is an effective solution to agent cooperation in heterogeneous and distributed environments. Nevertheless FIPA specification are not clear: Communication protocols: intra-platform and inter-platform; ACC and AMS: they have a strictly interaction and ACC is overburdened by the forward action; Default DF: its role and organisation are not clear. JEAP&FIPA

19. Loris PenseriniComputer Science Institute - University of Ancona 19 Project Overview DCBR approach

20. Loris PenseriniComputer Science Institute - University of Ancona 20 Query Representation: Case Memory Example: CBR approach

21. Loris PenseriniComputer Science Institute - University of Ancona 21 Local Query Rewriting: Terminology The problems (queries, Q) are classifies according to subsumption relation. In particularly, in this work, Q is decomposed and then classified. For example: CBR approach

22. Loris PenseriniComputer Science Institute - University of Ancona 22 Local Query Rewriting CBR approach Query Evaluation: the Mediator sends a decomposition of Q to the related information source and wait for their answers. Local Failure in Query Reuse: the Mediator is not able to rewrite Q since its case memory contains no past cases that can be used to do it. For example, it is the first time that the consumer formulates such a query, i.e., the consumer has a new information need. Local Failure in Query Evaluation: the Mediator sends a rewritten query to related sources and receives at least an empty answer. This means that the case memory of the mediator is not updated (an information source has been removed from system or changed its schema).

23. Loris PenseriniComputer Science Institute - University of Ancona 23 Distributed Query Rewriting Principal Cooperation Strategies: DCBR approach Partners Answers Queries Mediators Sources all, the newer, the older,... Original Rewritten Data the whole query, its components,...

24. Loris PenseriniComputer Science Institute - University of Ancona 24 A first strategy: the Mediator cooperates with other mediators, send them the original query, and asks for receiving the rewritten query. DCBR approach Theorem. Let M, N be two mediators such that M interacts with N when M fails. Let Cn(M) be the case memory of M after n interactions with N. Let Cn(N) be the case memory of N such that it does not change while N interacts with M. Then First consideration redundancy Distributed Query Rewriting

25. Loris PenseriniComputer Science Institute - University of Ancona 25 A first strategy: the Mediator cooperates with other mediators, send them the original query, and asks for receiving the rewritten query. DCBR approach Theorem. Let be n information sources. Let V be a view of. Let M, N be two mediators such that M interacts with N when M fails. Let Cn(M) be the mediated schema of M after n interactions with N. Then Second consideration Distributed Query Rewriting

26. Loris PenseriniComputer Science Institute - University of Ancona 26 A second strategy: the Mediator cooperates with information sources, sends them the original query, and receives the rewritten query. DCBR approach Theorem. Let be n information sources. Let V be a view of. V is represented as a case memory that does not change. Let M be a Mediator such that M interacts with when it fails. Let Cn(M) be the case memory of M after n interactions with. Then Distributed Query Rewriting

27. Loris PenseriniComputer Science Institute - University of Ancona 27 Notation Sol (Q) : the solution of Q (problem) respect the source S --> rewritten I (Sol (Q)) : the interpretation of the solution of Q both respect the source S --> data S SS Some Considerations DCBR approach

28. Loris PenseriniComputer Science Institute - University of Ancona 28 Mediator Prototype - Its language is a restrict set of Sequel Query Language - It implements a wide set of cooperation strategies - It collaborates with other mediators and ‘simple’ wrappers - Its Solution-Analysing and the Case-Memory Updating phases are still user driven

29. Loris PenseriniComputer Science Institute - University of Ancona 29 Case Memory Organization Mediator Prototype PROB:select EmployeeInfo from Employee SOL:"select EmployeeName from Employees@wrapper1@ iiop:// 129.100.27.30:900/acc" and "select EmployeeAddr from Address@wrapper2@iiop://193.205.128.40:1000/acc"

30. Loris PenseriniComputer Science Institute - University of Ancona 30 Failure in Retrieval - in reuse - in evaluation (partial or total) - by the user feedback Different cases of failure during retrieval: Mediator Prototype

31. Loris PenseriniComputer Science Institute - University of Ancona 31 Session Example Wrapper 1Wrapper 2 5.3: data 4.1: 2: It checks in its knowledge OK  Wrapper interrogation Agent iMediator 5.1: data 7.3: data 3.1: request to wrapper 1 3.2: request to wrapper 1 3.3: request to wrapper 2 4.2: 4.3: solve 7.1: data 5.2: failure 7.2: failure 6: Case Memory update 1: solution requests to mediator solve Mediator Prototype partial failure in evaluation failure in reuse user feedback

32. Loris PenseriniComputer Science Institute - University of Ancona 32 select EmployeeInfo from Employee Cooperation Strategies Mediator Prototype

33. Loris PenseriniComputer Science Institute - University of Ancona 33 Mediator 1 Mediator 2 W1 W2 W3 1: request 3: data 2: data 2: rewrite 3: data Cooperation Example Mediator Prototype

34. Loris PenseriniComputer Science Institute - University of Ancona 34 Strategies Evaluation The tests have evaluated, for each strategy, these kinds of problems : - knowledge growing - knowledge redundancy - response time - network overload Mediator Prototype

35. Loris PenseriniComputer Science Institute - University of Ancona 35 Strategies Evaluation The principals tests consist in five general queries (problems) all performed on the following five different strategies: Strategy 1: a mediator (M) sends the original query (Q) to another mediator (N) and requests for a rewrite Strategy 2: M sends a decomposition of Q to N and requests for a rewrite Strategy 3: M sends to N only the rewrited components of Q that it cannot solve and asks for the data Strategy 4: M directly sends a decomposition of Q to the wrappers and asks them for the data Strategy 5: M directly sends only the rewrited components of Q to the wrappers and asks for the data Mediator Prototype

36. Loris PenseriniComputer Science Institute - University of Ancona 36 Strategies Evaluation: Results Mediator Prototype

37. Loris PenseriniComputer Science Institute - University of Ancona 37 Strategies Evaluation: Results Mediator Prototype

38. Loris PenseriniComputer Science Institute - University of Ancona 38 Steps: principal phases of a communication (without any FIPA protocols) Service Messages: they do not contain data (query responses) inside otherwise they should be considered ‘Data Messages’ Small/Big Messages: it is another possible type of messages subdivision based on the content of a FIPA message (e.g. an ‘agree’ message is considered ‘small’, whereas an ‘inform’ messages is big, and so on) Strategies Evaluation: Results Mediator Prototype

39. Loris PenseriniComputer Science Institute - University of Ancona 39 Future Development - The designing and developing of a Mediator agent that uses a CBR engine based on DL - The designing of protocols that automatically choice the best strategy in response to a particular failure - To consider other useful cooperation strategies - To apply this architecture in a real case of interest to test its capacity (e.g. some kind of web applications)

40. Loris PenseriniComputer Science Institute - University of Ancona 40 References - M. Panti, L. Penserini, L. Spalazzi, S. Valenti, "A FIPA Compliant Agent Platform for Federated Information Systems", International Journal of Computer & Information Science, R. Y. Lee and H. Fouchal (eds), ACIS, 2000. - C. Cioffi, M. Panti, L. Penserini, L. Spalazzi, E. Tonucci, S. Valenti, ``An Agent-Based Platform for Federated Information Systems: Some Design Issues'', in Proc. of the International Conference on Software Engineering Applied to Networking and Parallel/Distributed Computing (SNPD '00), Reims, France, May 18-21, 2000. - M. Panti, L. Penserini, L. Spalazzi, ``A Critical Discussion about an Agent Platform based on FIPA Specification'', in Atti dell'Ottavo Convegno Nazionale su Sistemi Evoluti per Basi di Dati (SEBD 2000), L'Aquila Italia, 26-28 Giugno 2000. - M. Panti, L. Spalazzi, A. Giretti, “A case-based approach to information integration”, In Proceedings of the 26th International Conference on Very Large Databases, Cairo, Egypt, 10–14 September 2000. - Foundation for Intelligent Physical Agents, Fipa 2000 Specification. Geneva, Switzerland, October 2000. Available at: http://www.fipa.org/ - E. Plaza, J. L. Arcos, and F. Martìn, “Cooperative Case-Based Reasoning”, In G. Weiss, editor, Distributed Artificial Intelligence meets Machine Learning, Lecture Notes in Artificial Intelligence, Berlin, 1997. Springer Verlag. - G. Wiederhold, “Mediators in the architectures of future information systems”, in IEEE Computer, vol. 25, N. 3, pp. 38--49, 1992.