Abstract
This paper examines a decentralized and self-organizing approach inspired by ant behavior for building an information system in which metadata related to Grid resources is disseminated and logically organized. Each agent/ant, by analyzing its own past activity, copies and moves resource metadata among Grid hosts and contributes to collect resources belonging to the same class in a restricted region of the Grid, so decreasing the system entropy. A semi-informed resource discovery protocol exploits the ants' work: asynchronous query messages issued by clients are driven towards “representative peers” which maintain information about a large number of resources having the required characteristics. Agents control their activities, and query messages travel the network, according to self-organizing mechanisms based, respectively, on sematectonic and marker-based stigmergy, with no information about the global system state. Simulation analysis suggests that the combined use of the proposed resource mapping protocol (ARMAP) and resource discovery protocol (ARDIP) is profitable: as resources are progressively reorganized by the ARMAP process, users are able to find more and more results in a smaller amount of time.