Distributed swarms

Starting at the begin­ning of the new cen­tu­ry, the peer-to-peer (P2P) para­digm for buil­ding distri­bu­ted appli­ca­tions has attrac­ted the atten­tion of indu­stry and media, pri­ma­ri­ly due to the enor­mous suc­cess of systems such Napster and Gnutella fir­st, and Bittorent later. The mul­ti­tu­de of pro­jec­ts that have been pro­po­sed based on P2P need to con­front with com­mon pro­blems inclu­ding secu­ri­ty, relia­bi­li­ty and rou­ting. For the­se rea­sons, the IT com­mu­ni­ty is pro­mo­ting seve­ral ini­tia­ti­ves about P2P, such as the Intel’s Peer-to-Peer Working Group (for the deve­lo­p­ment of new P2P stan­dards) and the Sun’s JXTA pro­ject (for the deve­lo­p­ment of an open-sour­ce infra­struc­tu­re).

Unfortunately, none of the­se efforts seems to be aimed at sup­por­ting the scien­ti­fic inve­sti­ga­tion of the pro­per­ties of P2P systems. The Anthill pro­ject was an attempt to fill this hole, by pro­vi­ding a fra­mework hel­ping resear­chers in the stu­dy, the desi­gn and ana­ly­sis of P2P systems.

In order to pur­sue this goal, we have adop­ted the ant colo­ny para­digm. In this agent-based approach, arti­fi­cial ants of limi­ted indi­vi­dual capa­bi­li­ties move across a net­work of nodes try­ing to sol­ve a par­ti­cu­lar pro­blem. While moving, they build (par­tial) solu­tions and modi­fy the pro­blem repre­sen­ta­tion by adding col­lec­ted infor­ma­tion. The resul­ting system may be defi­ned as com­plex adap­ti­ve: indi­vi­dual ants are unin­tel­li­gent and have no pro­blem sol­ving capa­bi­li­ty; never­the­less, ant colo­nies mana­ge to per­form seve­ral com­pli­ca­ted tasks.

Results con­cer­ning file sha­ring and load balan­cing have been pre­sen­ted in [ICDCS02]  and [AP2PC03], respec­ti­ve­ly. A gene­ral over­view of the idea can be found in [LNCS02]  

The deve­lo­p­ment of Anthill has been inter­rup­ted. Yet, the expe­rien­ce gai­ned throu­gh it has been fun­da­men­tal to start a new pro­ject, cal­led BISON, that has been foun­ded by the European Community.

References

[ICDCS02]   Ozalp Babaoglu, Hein Meling, and Alberto Montresor. Anthill: A fra­mework for the deve­lo­p­ment of agent-based peer-to-peer systems. In Proc. of the 22th Int. Conference on Distributed Computing Systems (ICDCS’02). IEEE, Vienna, Austria, July 2002. [PDF][Bibtex].

[LNCS02]   Alberto Montresor, Hein Meling, and Ozalp Babaoglu. Towards adap­ti­ve, resi­lient and self-orga­ni­zing peer-to-peer systems. In Web Engineering and Peer-to-Peer, num­ber 2376 in Lecture Notes in Computer Science, pages 300–305. Springer-Verlag, 2002. [PDF][Bibtex].

[AP2PC03]   Alberto Montresor, Hein Meling, and Ozalp Baboglu. Messor: Load-balan­cing throu­gh a swarm of auto­no­mous agen­ts. In Proc. of the 1st Workshop on Agent and Peer-to-Peer Systems (AP2PC’02), num­ber 2530 in Lecture Notes in Artificial Intelligence, pages 125–137. Springer-Verlag, Bologna, Italy, July 2003. [PDF][Bibtex].