All your network are belong to us: a transport framework for mobile network selection

Mobile devices come with an assortment of networks: WiFi in two different frequency bands, each of which can run in infrastructure-mode, WiFi-Direct mode, or ad hoc mode; cellular radios, which can run in LTE/4G, 3G, or EDGE modes; and Bluetooth. But how should an app choose which network to use? There is no systematic solution to this problem today: in current practice the choice is almost always left to the user, who usually has no idea what's best. In fact, what's best for a user depends on the app's performance objectives (throughput, delay, object load time, etc.) and the user's constraints on cost and battery life. Besides, what's best for a single user or app must be balanced with what's best for the wireless network as a whole (individual optimality vs. social utility). This paper introduces Delphi, a transport-layer module to resolve these issues. Delphi has three noteworthy components: "local learning", in which a mobile device estimates or infers useful properties of different networks efficiently, "property sharing", in which mobile devices share what they learn with other nearby devices, and "selection", in which each node selects a network using what it has observed locally and/or from its neighbors.