Towards Near Optimal WiFi Offloading With Uncertain Contact Duration

Due to the simplicity of implementation, user-initiated Wi-Fi offloading becomes more and more popular, and naturally the benefits of users become the main optimization goal. We notice the intercontact and intra-contact durations could be uncertain in reality by reason of the user mobility and network dynamics. The two uncertain durations can cause great impact on the benefit of users; however, they were either ignored or simply assumed to be deterministic in most previous works. In this paper, for the first time, we study Wi-Fi offloading problem with uncertain contact durations. The aim is to guarantee the benefit of users (delay and payment) without damaging operator's benefit (amount of the offloaded traffic) at the same time. We propose a multi-armed bandit (MAB)-based online offloading scheme (MABOO) to solve the problem and prove the near-optimality of MABOO in terms of the utility theoretically. Extensive simulations show that MABOO always approaches the optimal scheme and achieves higher utility as well as offloads more traffic compared with the minimal payment and on-the-spot-offloading schemes.