Sizing Up User Traffic: Flow-based Mobile Data Offloading Over WiFi

We propose a smart offloading policy that dynamically assigns data flows to the WiFi and cellular interfaces, so as to minimize a given cost function (related to energy consumption and cellular plan usage), while keeping the average per-flow delay bounded. The basic insight of the proposed Threshold Policy is to assign larger flows to the network that provides the best rate (often WiFi), and smaller flows to the other, since energy is generally related to the time needed to send/receive data. However, choosing the size cutoff optimally must also consider load-balancing and queueing aspects, WiFi availability, flow size statistics, and user/application preferences. We validate our model against simulations, and show that our policy outperforms other standard or smart policies, achieving considerably better energy-delay trade-offs, while only offloading a small percentage of (large)flows. Initial measurements performed on an Android-based offloading prototype further support our findings.