Intelligent network selection and energy reduction for mobile devices

Abstract

The popularity of mobile devices has stimulated rapid progress in both Wi-Fi and cellular technologies. Before LTE was widely deployed, Wi-Fi speeds dominated cellular network speeds. But that is no longer true today. In a study we conducted with a crowd-sourced measurement tool used by over 1,000 users in 16 countries, we found that 40% of the time LTE outperforms Wi-Fi, and 75% of the time the difference between LTE and Wi-Fi throughput is higher than 1 Mbits/s. Thus, instead of the currently popular "always prefer Wi-Fi" policy, we argue that mobile devices should use the best available combination of networks: Wi-Fi, LTE, or both. Selecting the best network combination, however, is a challenging problem because: 1) network conditions vary with both location and time; 2) many network transfers are short, which means that the decision must be made with low overhead; and, 3) the best choice is determined not only by best network performance, but also constrained by practical factors such as monetary cost and battery life. In this dissertation, we present Delphi, a software controller for network selection on mobile devices. Delphi makes intelligent network selection decisions according to current network conditions and monetary cost concerns, as well as battery-life considerations. Our experiments show that Delphi reduces application network transfer time by 46% for web browsing and by 49% for video streaming, compared with Android's default policy of always using Wi-Fi when it is available. Delphi can also be configured to achieve high throughput while being energy efficient; in this configuration, it achieves 1.9 x the through-put of Android's default policy while only consuming 6% more energy. Delphi improves performance but uses the cellular network more extensively than the status quo, consuming more energy than before. To address this problem, we develop a general method to reduce the energy consumption of cellular interfaces on mobile devices. The key idea is to use the statistics of data transfers to determine the best times at which to put the radio in different power states. These techniques not only make Delphi more useful in practice but can be deployed independently without Delphi to improve energy efficiency for any cellular-network-enabled devices. Experiments show that our techniques reduce energy consumption by 15% to 60% across various traffic patterns.