One of open issues in grid computing is efficient resource discovery. Centralized solutions have been proved inefficient for global scale resource discovery. In addition, due to lacking of semantic support, the discovery mechanisms of current solutions do not have the flexibility to perform flexible resource matches for requirements of tasks or jobs. It is thus desirable if we could have a decentralized discovery mode, which does not need global knowledge but local knowledge only and has a semantic matching capability to perform flexible resource matches and decide the degree of flexibility that the system is granted.
In this talk, by introducing semantic similarity of domain ontology and using decentralized mechanism of finding neighbours, we propose a heuristic algorithm. Without overhead of negotiation, the algorithm allows individual resource agents to semantically interact with neighbour agents based on local knowledge and dynamically form a resource service chain to complete a task (the resource agents only need know the information of its neighbour). The algorithm ensures resource agent's ability to cooperate and coordinate on neighbour knowledge requisition for flexible problem solving. The developed algorithm was evaluated by investigating the relationship between the success probability of resource discovery and semantic similarity under different factors including task complexity, task loading rate, and network scale etc. The experiments show our algorithm could flexibly and dynamically discover resources and therefore provide a valuable addition to the field.