We implemented AntHocNet, a novel algorithm for routing in mobile ad hoc networks (MANETs). AntHocNet is designed to show adaptive and robust behaviour with respect to network changes. This is achieved by means of a hybrid design that combines reactive and proactive behaviours allowing to both anticipate and respond in timely fashion to sudden disruptive events. At the very core of the algorithm there are two adaptive learning mechanisms: the Monte Carlo sampling and learning typical of ant-based approaches, and an information bootstrapping process typical of many reinforcement learning schemes. Operating at different timescales the two mechanisms allow to continuously adapt nodes' routing tables and efficiently set-up multiple routing paths optimised with respect to a number of metrics of interest, such as delay, throughput, signal-to-noise ratio, etc.
Using QualNet, a popular commercial network simulator, AntHocNet performance has been assessed through extensive simulation studies considering a number of open space and urban/structured MANET scenarios with different traffic and mobility dynamics. Compared to other state-of-the-art algorithms such as AODV and OLSR, AntHocNet always shows superior performance and robustness of response over the different scenarios, at the expenses of a comparable or even lower routing overhead. AntHocNet also shows excellent superior performance in wireless mesh networks and in wired scenarios (outperforming OSPF). All these results support the view that AntHocNet has an enormous potential for use in real-world MANETs, and, more in particular, in heterogeneous mesh networks composed of mobile, fixed, and wired nodes. In fact, the hybrid and composite nature of its design makes it an algorithm able to cope equally well with a number of different networks and network dynamics, as well as with a number of different modes/characteristics coexisting in the same, possibly heterogeneous, network.
We are proceeding to an implementation of the algorithm on Linux-based wireless/wired devices. In a few months we should have the first data from real-world experiments in a mesh network. If real-world results will confirm the simulation ones we will proceed considering the deployment of mesh networks based on the use of AntHocNet.