DIET: Decentralised Information Ecosystem Technologies Paul Marrow Intelligent Systems Laboratory

DIET: Decentralised Information Ecosystem Technologies

Paul MarrowIntelligent Systems Laboratory

© British Telecommunications plc, 2003 DIET @ UIE 2

DIET: Some history

At BT, involved in development of applications drawing inspiration from nature since (at least) 1995

Common interest with other partners in developing software systems bringing together many interacting entities

Hence, when European Commission declared interest in information ecosystems… DIET


Universal Information Ecosystems

What was said in the call: Information Ecosystem

- vision of future information devices interacting in many complex ways akin to natural ecosystems

- populated by infohabitants devices, virtual entities software agents individuals organisations


The Real Thing


Why DIET?

Decentralised- run applications made of many distributed entities

(mobile agents)- collaboration without need for centralised control

Information Ecosystem- manage information through interaction of many

entities- analogous to interactions in natural ecosystems

Technologies- basis for deployment of distributed technologies


Involvement and roles

BTexact Technologies

Technical University of Crete

DFKI

Universidad Carlos III de Madrid

Visualisation

Economic interaction

Information brokering

Information filtering and mining

Core platform

Applications

Coordination


Progress

WP1: The DIET Platform WP2: Information Alert WP3: Information Brokering WP5: Visualisation WP6: Dissemination and Exploitation


The DIET platform> architecture

Applicationlayer

ARC layer

Core layer

Applicationreusableservices

Applicationcomponents

Visualisationcomponents

Validationcomponents

DIET platform kernelDebugging &visualisationkernel

Debugging &visualisationcomponents


DIET platform kernel> overview

The “physics” of the DIET ecosystem Hierarchy of elements

- Universe, World, Environment, Infohabitant (=Agent), Connection, Message

DIET environment provides:- Agent creation- Agent destruction- Agent migration- Agent communication


Lightweight agents> infohabitants

Agents can be very lightweight: Need only one thread...

- Event portal

… or even no thread!- Thread sharing

Connection contexts avoid look-up tables 500000 Agents on single machine.


Thread sharing

ThreadPool

Environment

Environment

Environment

ThreadPool


Decentralisation

Architecture fully decentralised No central agent naming scheme

- Agents have identities assigned locally- Identities are randomly initialised bitstrings

No central world registry- World neighbourhoods defined in P2P fashion- Connections between worlds created on demand


The platform: further directions

Basis for research within DIET Basis for application development within

BTexact Scenarios for third-party licensing under

investigation

See Marrow et al. (2001); Hoile et al. (2002)


Information Retrieval, Filtering and Mining

Universidad Carlos III de Madrid- with Technical University of Crete

Used DIET platform as basis to build societies of agents engaged in information push and pull tasks

Can use these agents to construct information retrieval, filtering and mining applications with which to validate the performance of the system


I-Gaia: the world of Infocytes

TI

TI

TIMI

MI

MIMI

see Gallardo-Antolin et al. (2002)

SI


Information Brokering

Technical University of Crete- together with BTexact Technologies- and Universidad Carlos III de Madrid

Models for the brokering of information in information ecosystem middleware

Implementation of brokering systems based on DIET platform

Development of self-organising communities application


An Open Information Ecosystem

MiddlewareMiddleware

usersInformationconsumers

information sources

Informationproducers

queries/profiles

information

information

information


Visualisation

DFKI (Deutsches Forschungszentrum für Künstliches Intelligenz)

Simple visualizer of multi-agent system constructed early in project

Later work developing more complex user and development interface

See van Lengen & Bähr (2002)


Visualisation> architecture

Applicationlayer

ARC layer

Core layer

Applicationreusableservices

Applicationcomponents

Visualisationcomponents

Validationcomponents

DIET platform kernelDebugging &visualisationkernel

Debugging &visualisationcomponents


Visualisation


Visualisation> further directions

Provision of different views of DIET multi-agent system

Interface to allow visual programming of DIET systems

Different approaches to debugging of DIET applications


Dissemination and Exploitation

Dissemination- all partners

see e.g. references

Exploitation- BTexact Technologies leads application

development- Implementation of experimental scenarios- Development together with other projects

DIET providing infrastructure


Examples

Self-organising communities Evolving preferences SWAN and DIET


Self-organising communities> ideas

Flexible and effective organisation Organised by middle agents Users’ expressed characteristics

- track user preferences/interests from user behaviour

- no difficult user preference representation or matching problems

Flexible, scalable, robust, distributed


Self-organising communities> implementation

User agents- register/unregister with a middle agent- initiate a query/process query results

Middleman agents- search scheme- award scheme- exchange scheme


Formation of communities


Self-organising communities> results

Algorithm is highly scalable Improves speed of search for users Now basis for further application

development- see Wang (2002)


Evolving group preferences

Facilitating interaction between users in different environments

Using scout agents to locate alternative environments for communication

Selecting for improved adaptation to environments

Combining agent population with evolutionary algorithm

Using evolutionary algorithm to evolve preferences

See Marrow et al. (2002)


Scout SuccessSingle Category

8 Users

128 Users


Environment PreferenceSingle Category

8 Users

128 Users


Changing scout success


SWAN: Nodes in p2p system

Physical network

522

921 773

391

13.10.2.7

13.10.2.4

25.2.12.4

25.10.13.5

18.6.5.15494

identity (location-independent)address


SWAN: Node look-up for p2p

Problem: identity of node address of node?

Peer-to-peer (p2p) domain requires:- fully decentralised system - yet robust and scalable

Solution: - Let nodes self-organise into a Small World

Network (SWN)- Use SWN to find nodes


Complementary Technologies

DIET - fail-fast- fully decentralised- indefinitely scalable, but no global addressing

SWAN- failure tolerant- fully decentralised- highly scalable global addressing system

See Bonsma & Hoile (2002)


Discussion

Activities in DIET covering a range of research directions

Information ecosystem as implemented does not stick closely to biological inspiration

But provides useful infrastructure for a variety of experiments and applications

How will this be used in the future? What links are possible with results of other

projects?


Discussion

How can results from information ecosystems projects contribute to the emergence of...

- peer-to-peer computing- ad-hoc mobile computing- ubiquitous computing- pervasive computing

How integrate these areas with agent technologies in information ecosystems?


References Marrow, P, M. Koubarakis, R.H. van Lengen, F. Valverde-Albacete, E. Bonsma, J. Cid-Suerio, A.R.

Figueiras-Vidal, A. Gallardo-Antolín, C. Hoile, T. Koutris, H. Molina-Bulla, A. Navia-Vázquez, P. Raftopoulou, N. Skarmeas, C. Tryfonopoulos, F. Wang, C. Xiruhaki (2001) Agents in Decentralised Information Ecosystems: The DIET Approach. In: Proceedings of Symposium on Intelligent Agents for E-commerce, AISB01 (2001 Artificial Intelligence and Simulation of Behaviour) Convention, University of York, April 2001.

Bonsma, E. & Hoile, C. (2002) A distributed implementation of the SWAN peer-to-peer look-up system using mobile agents. In: Proceedings of International Conference on Autonomous Agents and Multi-Agent Systems 2002 (AAMAS2002), tba, Bologna, July 2002.

Gallardo-Antolín, A., Navia-Vázquez, A., Molina-Bulla, H.Y., Rodríguez-González, A.B, Valverde-Albacete, F.J., Cid-Sueiro, J., Figueiras-Vidal, A.R., Koutris, T., Xirouhaki, C. & Koubarakis, M. (2002) I-Gaia: an Information Processing Layer for the DIET Platform. In: Proceedings of International Conference on Autonomous Agents and Multi-Agent Systems 2002 (AAMAS2002), pp. 1272-1279, Bologna, July 2002.

Hoile, C., Wang, F., Bonsma, E. & Marrow, P. (2002) Core specification and experiments in DIET: a decentralised ecosystem-inspired mobile agent system. In: Proceedings of International Conference on Autonomous Agents and Multi-Agent Systems 2002 (AAMAS2002), pp. 623-630, Bologna, July 2002.

van Lengen, R.H. & Bähr, J.-T. (2002) Visualisation and debugging of decentralised information ecosystems. In: Proceedings of Dagstuhl Seminar on Software Visualization, Dagstuhl, February 2002. Springer, Berlin.

Marrow, P., Hoile, C., Wang, F. & Bonsma, E.R. (2002) Evolving preferences among emergent groups of agents. In: Proceedings of Symposium on Adaptive Agents and Multi-Agent Systems (AAMAS-II), AISB02 (2002 Artificial Intelligence and Simulation of Behaviour) Convention, Imperial College, London, April 2002.

Wang, F. (2002) Self-organising communities formed by middle agents. In: Proceedings of International Conference on Autonomous Agents and Multi-Agent Systems 2002 (AAMAS2002), pp. 1333-1339, Bologna, July 2002.


Acknowledgements

BTexact DIET team- Erwin Bonsma, Cefn Hoile, Fang Wang

Universidad Carlos III de Madrid- Francisco Valverde-Albacete et al.

Technical University of Crete- Manolis Koubarakis et al.

Deutsches Forschungszentrum für Künstliches Intelligenz

- Rolf van Lengen et al.

Documents

DIET: Decentralised Information Ecosystem Technologies Paul Marrow Intelligent Systems Laboratory