Towards an Adaptive Grid Scheduling: Architecture and Protocols Specification

Grid is known to be a heterogeneous, distributed and dynamic environment. In order to take fully advantages from grid power, Grid scheduling must take into consideration the environment's constraints and be adaptive. In this work, Grid architecture is fully rethought in terms of agents ill order to implement a cooperative and adaptive scheduling. At a macro level, our architecture enables flexible cooperation among its Components using high level interaction protocols. At the micro level, agents in charge of scheduling perform all adaptive behaviour since they are able to perceive their environment and its disturbances, to reason and to deliberate about the actions to undertake in order to adapt. This is made possible by the use of Belief-Desire-Intention mechanisms. For that purpose, we propose a conceptual model useful for the perception function. Also, a typology of adaptive rules useful for the deliberation step is given. Component's behaviour are specified and simulated with Petri-Nets.