engleski

Traffic-Dependent Routing Based on Self-adaptation

This paper presents a method of self-adaptive routing, considered as an intelligent agent support for client applications. The method is based on the reinforcement Q-learning of an immobile software agent. In this model the control is sensitive to the current availability of the network links. Inspiration is found in the paradigm of software agent technology and some of its possible repercussions that we suppose to appear when applied in telecommunication network. The main property of the usage of an intelligent agent is the opportunity to permanently transfer adaptation for the jobs that have to be sent from a client to the server side. The search for optimal path is based on continuous monitoring of available channel capacity between the client and the network nodes. The actions that follow monitoring are focused on the selection of optimal throughput via the node that enables maximal exploitation of the channel. In some situations information flow between the client and server can be significantly reduced as a consequence of traffic load, but also of the throughput choice. The results of our simulations show that the exploration and usage of knowledge about throughput properties in the network can optimally utilize redundant capacities.

intelligent agents; routing; reinforcement

Frontiers in Artificial Intelligence and Applications = FAIA Series (0922-6389) ; jedna od podsekcija: Knowledge-Based Intelligent Engineering Systems