Today's computational grids are using the standard IP routing functionality, that has basically remained unchanged for 2 decades, considering the network as a pure communication infrastructure. With the grid's distributed system point of view, one might consider to extend the commodity Internefs basic functionalities. Higher value functionalities can thus be offered to computational grids. In this paper, we report on our early experiences in building application-aware components and in denning an active grid architecture that would bring the usage of computational grid to a higher level than it is now (mainly batch submission of jobs). To illustrate the potential of this approach, we first present how such application-aware components could be built and then some experiments on deploying enhanced communication services for the grid. We will show how reliable multicast and QoS mechanisms could deploy specific services based on the grid application needs. \textcopyright 2004 Elsevier B.V. All rights reserved.