Experimental measurement of the capacity region of wireless networks

We present a method for experimentally measuring the capacity region of a wireless network, defined here as the set of all possible combinations of simultaneously achievable link transmission rates. Due to the inherent complexity of the problem, measuring the capacity region exactly typically involves a prohibitively large volume of experiments. Therefore, the method is based on a judicious, tunable algorithm, employing online machine learning, which aims at reducing the volume of experiments conducted. In order to evaluate the efficiency of the method, and also to demonstrate the usefulness of having the capacity region available, we apply our method to a network of Raspberry Pi nodes placed inside a building and communicating using IEEE 802.11. In particular, we compare the performance of an optimal, centrally organized communication scheme, computed using the measured capacity region within a network optimization formulation, with the performance of a simple communication scheme that uses multihop TCP flows.