Åóóóðð Áòøøöñññññööö× Ëùôôóöøøòò Áòòóöññøøóò Ëëëööòò Øûò Åóóóðð Í××ö× Aeóöððþþ Òò Òò Äù Åóöööù

ION LAYER SHADOW MOBILE DEVICE AGENT Fig. 8. Appli ation Intera tions The appli ation intera tions are illustrated in Figure 8. A mobile user uses a browser on a laptop to a ess information. The browser sidebar intera ts dire tly with a User Appli ation, whi h is also an agent running on the laptop. User's requests are sent by the browser sidebar to the User Appli ation. If no User Agent is started on the xed infrastru ture, the User Appli ation requests the abstra tion layer to start a User Agent. Then, the User Appli ation is ready to route user's requests to the User Agent using the delivery me hanism we des ribed in Se tion 4. For a request to get re ommendations on related urls or to get other users' bookmarks, a \SimilarLinks" query is onstru ted. An inform \Bookmarks" message is reated if the User Appli ation re eived a request to export a set of bookmarks to the infrastru ture. These messages are then forwarded to the User Agent. For a \SimilarLinks" query, the User Appli ation would get a set of urls to similar do uments or a set of other users' bookmarks in return. These urls or bookmarks are then forwarded to the browser sidebar to be displayed. On the infrastru ture, a User Agent queries a Registry Agent for a Re ommender Interfa e Agent, with whi h it intera ts in order to use servi es o ered by the Re ommender system. Every \SimilarLinks" query or \Bookmarks" inform message re eived from the mobile devi e is forwarded to the Re ommender Interfa e Agent. Results of \SimilarLinks" queries are returned to the mobile devi e. Intera tions between a User Appli ation on the mobile devi e and a User Agent on the xed infrastru ture are supported by our abstra tion layer. At this stage, we have not ompleted a formal evaluation of our ar hite ture, but we are olle ting observations about it. Given the appli ation, we informally ompared the use of a laptop with our abstra tion layer and without it. In both ases, the laptop an intera t with appli ations on the xed infrastru ture. In the se ond ase, when the laptop is dis onne ted, messages may be lost and resending of messages will has to be programmed at the appli ation level. In the 15 rst ase, the use of a mobile agents and the transparent routing of messages to them [8℄ solved the problem of delivering messages to the laptop; alternatively in the se ond ase, full IPv6 may be ne essary to route messages to the laptop mobile address. Finally, by addressing the problem of the reliable delivery and of the routing of messages in an abstra tion layer, we have designed a generi solution reusable by other appli ations having to support mobile users. 6 Related Work In [6℄, mobile agents are used to move between resour es on the xed infrastru ture to take advantages of those resour es in order to a omplish tasks for a mobile user. An Agent Gateway, whi h is a stationary host, is a ting as the mediator between the wireless devi e and xed infrastru ture resour es. This is di erent from our approa h as we adopt a mobile agent to be su h a mediator. Having a mobile mediator is more exible as it an move loser to the urrent lo ation of the user, whi h allows lo al ommuni ation to be established. The \Personal Agent System" [1℄ provides a mobile user with a personalised information retrieval servi e. The Personal Agent is a mobile agent that resides on the xed infrastru ture and ommuni ates with agents residing on the mobile devi e. The system is similar to ours in the sense that it involves migration of the Personal Agent to other stationary servers so that it follows the mobile user around in the wired network, while the user moves around in the wireless network. In omparison, our abstra tion layer provides more exibility sin e we allow multiple mobile agents to exist when the user's urrent lo al network is not onne ted with the user's previous lo ation. The Mobile Agents Platform (MAP) ar hite ture [2℄ involves data servers to store results a quired by a mobile agent for a mobile user on e the mobile user is dis onne ted from the network. When re onne ted, the user has to undergo multiple ommuni ation steps to get the result, like having to query the lookup table for data server address and then to query the data server for the result. Our approa h is mu h simpler sin e we provide a store and forward me hanism built in the abstra tion layer, whi h allows the results of user's queries to be forwarded to the mobile user on e the mobile user is re onne ted to the network. In the M-Commer e Framework [7℄, a mobile agent alled Servi e Agent is moving around the wired network to gather information for a mobile user, while another mobile agent alled Courier Agent is migrating to the mobile devi e to establish an intera tion with the Servi e Agent on the xed infrastru ture. Migrating the Courier Agent to the mobile devi e in order to intera t with the Servi e Agent puts more burden on the network onne tion than a migration between two hosts on the xed network, as it involves the ability to move the agent state and ode, whi h in ludes the serialisation and deserialisation of the transferred data through the low bandwidth wireless ommuni ation hannels. In our abstra tion layer, we adopt a simpler approa h, where an appli ation residing on the mobile devi e is responsible for intera ting with appli ations on the xed infrastru ture. 16 In the Ta oma Ar hite ture [4℄, a support spe i to PDA appli ation is provided using an entity alled \hostel", whi h is the host that a PDA normally uses to syn hronise data with. The hostel is also assumed to a t as the network provider or proxy for the PDA, i.e. the hostel is a networked workstation. In this ar hite ture, mobile agents are used to gather information on the wired network assuming the presen e of a host that they an inquire in ase the PDA is not onne ted. This approa h is suitable for a PDA user that has the hostel as the only onne tion point needed for the PDA. But in the ase of a user who is always on the move and needs to onne t to di erent hosts, a more exible approa h su h as having a mobile agent a ting as the \hostel" is more suitable. 7 Con lusion A mobile agent able to migrate around the network trying to stay as lose as possible to the mobile user, gives a major advantage by allowing lo al ommuni ation to be established with the mobile devi e. With this apability, the mobile agent is designed to be the main omponent in our abstra tion layer, whi h allows transparent intera tions between xed infrastru ture appli ations and appli ations on mobile devi es. This paper has presented an appli ation, whi h supports information sharing between mobile users in virtual meeting rooms. The main hallenge in developing the appli ation is to onstru t an intermediary layer, whi h supports seamless ommuni ation between a traveling mobile user and virtual meeting rooms hosted by the xed infrastru ture. We have introdu ed an ar hite ture and algorithm of the intermediate layer, based on a mobile agent alled Shadow. This intermediary layer takes are of oordination of multiple Shadows, as well as the ommuni ation between a mobile devi e and its Shadows. It is de ned as an abstra tion layer, whi h hides the details of ommuni ation and oordination, allowing transparent intera tions between xed infrastru ture appli ations and appli ations on a mobile devi e. Having the intermediary layer has made the implementation of the appli ation straightforward, in whi h the layer takes are of omplex intera tions with mobile devi es. An agent-based Re ommender system [11℄ is used in our appli ation to provide an information sharing environment between User Agents, whi h are mobile-users' representative in the virtual meeting room. The User Agents intera t transparently with a mobile devi e through the intermediary layer. We believe su h ability is important to allow more appli ations for mobile users to be easily developed. 8 A knowledgement This resear h is funded in part by QinetiQ and EPSRC Magnitude proje t (referen e GR/N35816). 17 Referen es 1. Debbie Chyi. An Infrastru ture for a Mobile-Agent System that Provides Personalized Servi es to Mobile Devi es. Te hni al Report TR2000-370, Dartmouth College Computer S ien e, 2000. 2. A. La Corte, A Pulia to, and O. Tomar hio. An Agent-based Framework for Mobile Users. In Pro eedings of European Resear h Seminar on Advan es in Distributed Systems 1999, Madeira, Portugal, 1999. 3. Robert S. Gray, David Kotz, Ronald A. Peterson, Joy e Barton, Daria Cha on, Peter Gerken, Martin Hofmann, Je rey Bradshaw, Maggie R. Breedy, Renia Je ers, and Niranjan Suri. Mobile-Agent versus Client/Server Performan e: S alability in an Information-Retrieval Task. In Mobile Agents, pages 229{243, 2001. 4. Kjetil Ja obsen and Dag Johansen. Mobile Software on Mobile Hardware { Experien es with TACOMA on PDAs. Te hni al Report 97-32, Department of Computer S ien e,University of Troms, Norway, 1997. 5. Danny B. Lange and Mitsuru Ishima. Programming and Deploying Java Mobile Agents with Aglets. Addison-Wesley, 1998. 6. Q.H. Mahmoud. MobiAgent { An Agent-based Approa h to Wireless Information Systems. In Pro eedings of the 3rd International Bi-Conferen e Workshop on Agent-Oriented Information Systems (AOIS-2001), Montreal, 2001. 7. Patrik Mihailes u and Walter Binder. A Mobile Agent Framework for MCommer e. Computer S ien e 2001, GI/OCG annual Convention:2:959{967. . 8. Lu Moreau. Distributed Dire tory Servi e and Message Router for Mobile Agents. S ien e of Computer Programming, 39(2{3):249{272, 2001. 9. Lu Moreau. A Fault-Tolerant Dire tory Servi e for Mobile Agents based on Forwarding Pointers. In The 17th ACM Symposium on Applied Computing (SAC'2002) | Tra k on Agents, Intera tions, Mobility and Systems, Madrid, Mar h 2002. 10. Lu Moreau, Ni k Gibbins, David DeRoure, Samhaa El-Beltagy, Wendy Hall, Gareth Hughes, Dan Joy e, Sanghee Kim, and Danius Mi haelides. SoFAR with DIM Agents An Agent Framework for Distributed Information Management. In Pro eedings of the 5th International Conferen e on the Pra ti al Appli ation of Intelligent Agents and Multi-Agent Te hnology (PAAM 2000), pages 369{388, 2000. 11. Lu Moreau, Norliza Zaini, Jing Zhou, Ni holas R. Jennings, Yan Zheng Wei, Wendy Hall, David De Roure, Ian Gil hrist, Mark O'Dell, Sigi Rei h, Tobias Berka, and Claudia Di Napoli. A Market-Based Re ommender System. In Paolo Giorgini, Yves Lesp eran e, Gerd Wagner, and Eri Yu, editors, Pro eedings of the Fourth International Bi-Conferen e Workshop on Agent-Oriented Information Systems at AAMAS 2002 (AOIS'02), Bologna, Italy, July 2002. http://CEUR-WS.org/Vol59/. 12. Mark Weiser. Some Computer S ien e Problems in Ubiquitous Computing. Communi ations of the ACM, 36(7):74{84, July 1993.