Adapting the Overlay Network topology based on Traffic Matrix Estimation

The dynamic topology design is one of the important steps for the design of an overlay network. Traffic characteristic and volume, and behavior of nodes are the main issues that affect the performance of overlay networks, and have to be considered in designing a good topology. In this paper, we study the problem of designing the dynamic overlay topology based on estimating and predicting the Origin Destination OD pair flows. We consider the co operative behavior of the nodes in the network to minimize the overall cost of the network. The Kalman Filter prediction technique is used to predict the flows based on the data observations representing the number of packets on each link in the network. Then, we propose a heuristic to find nearoptimal overlay topologies with a reduced complexity. For a small-size network, we have applied the prediction technique including the heuristic to observe the change in the overlay topology with the change of the traffic demands between nodes. Results show that in most cases, the overlay topology is changed frequently for some changes of traffic demands. Also the far nodes in the network change the overlay topology frequently by adding and dropping overlay links between them.