Several digital libraries use the model of "harvest metadata" that is based on the Open Archive Infrastructure / PMH. On a certain number of sites, the model is satisfactory. In the context of a digital library collection, the performance of such infrastructure based on the discovery of informational resources and the execution of indexation requires that we move to an architecture based on high performance. High performance is a constraint becoming more creditworthy in grid computing and storage architectures. This presentation proposes a network architecture based on grid computing technology that allows parallel harvesting of metadata and supports harvest planning and its dynamic and uniform distribution between grid nodes, in order to grow the harvest performance of the basic OAI-PMH model while optimizing the harvest time and metadata processing.