1. Introduction

In the context of an increasingly decentralized and heterogeneous network such as the Internet today, it is very difficult for the applications to know how much quality they can really expect. A lot of research effort has been dedicated to techniques to offer QoS guarantees, but these techniques are only effective when deployed in all the nodes concerned by a communication, or at least in all the nodes where resource shortage may occur. However, it is very difficult to achieve global agreements such that these techniques can be deployed. They are feasible in a private enterprise network, ISP or isolated operator, but not at the global scale. Therefore the only realistic answer seems to be to rely on applications that are able to adapt to a wide range of network conditions in a dynamic way, and in particular to the amount of resources available when these resources cannot be reserved in advance.

An adaptive application must be able to make optimal use of the available resources, and be able to adapt itself to fluctuations in resource availability. But this kind of application faces the difficulty of obtaining enough information about the network conditions, due to the Internet black-box model.

Clearly the use of more intelligent network elements in the network can allow applications to obtain the feedback they need to perform the adaptation functions more easily. Recently router support has been considered to assist adaptive...