Ant routing vs. Q-routing in telecommunication networks

In our simulations we analyzed two classes of adaptive routing algorithms: Q-routing algorithms based on reinforcement learning techniques, and ant-routing based on swarm intelligence. We use a simulation environment to compare the performance of adaptive routing algorithms. Performance of adaptive algorithms was tested under a constant load, and also under various non-stationary loads, to find out how load time dynamics might interfere with learning dynamics. We tested step load changes and periodic load changes. We also tested an influence of non-homogeneous loads on the behavior of routing algorithms, motivated by Dos and DDoS network attacks. We investigate robustness to parameter changes, and adaptiveness to dynamically changing load levels and traffic patterns. We demonstrated that both classes of adaptive routing algorithms adapt well to occasional changes of the load and perform very well under periodic load level changes.