TABLE OF CONTENTS We have presented a survey of current research on agent and active network techniques applied to adaptive applications, with special attention to optimization and market-based approaches. We have also described a model for trading resources inside an active network node, which draws many elements from agent technology. We have applied the model to a concast audio mixing application which trades off link resources against memory in the presence of bottleneck links. The concast application is able to take congestion control decisions locally at each active node, such that no closed loop feedback between source and destination is needed. Using simulations, we studied three different strategies to make a decision on the amount of resources to use: two naive strategies based on the cheapest price, and a strategy that makes use of utility function weights. The results indicate that the first two strategies are already able to make improvements over the case when no congestion control is used, but they use resources inefficiently. The third strategy gives better results, achieving a stable and efficient sharing of resources.
We have several research directions to pursue: the most immediate one is to perform more complex simulations involving multiple node and link types, resource manager types, active and non-active nodes, different user strategies, etc. An implementation over a real active networking platform is also envisaged for the near future. We also plan to investigate the issues of dynamic resource manager upgrade with the help of mobile agents. The precise...
