TABLE OF CONTENTS In previous work [YAM 00a] we proposed a model for trading resources inside an active node, and applied it to an audio application. This paper is an extended, corrected and updated version of that work. The model was derived from our earlier work on a layered multicast protocol [YAM 00b], therefore it is not a mere abstract proposal but is motivated by concrete application needs.
Our model aims at offering a generic communication abstraction between active network agents, such that different adaptive applications and different resource management policies can be implemented. We are mainly concerned with adaptation to available resources when resources cannot be reserved in advance, either because the router itself does not support reservations, or because the network is heterogeneous and some of the routers along the path offer no QoS support.
In the model, two types of agents communicate to seek an equilibrium. Each agent tries to optimize its own benefits: on one side resource manager agents have the goal of maximizing resource usage while maintaining a good performance level. On the other side, user agents try to obtain a better quality/price relation for the resources consumed, and to efficiently manage their own budgets avoiding waste. Both types of agents are implemented as AAs with different privileges, and they communicate such that the resource managers can "sell" resources to the user agents at a price that varies as a function of the demand for the resource.
A currency is introduced into the system...
