Improving Phones through Surveillance
Apps that track how people use their phones could help make the devices more efficient.
A cell-phone application that logs everything the phone’s user does–from sending e-mail to playing games–may not sound so desirable. But researchers are deploying the software to see if they can determine the best ways to improve the battery life of phones and uncover network dead spots.
Working with colleagues at Microsoft Research, Hossein Falaki, a PhD candidate at UCLA’s Center for Embedded Network Sensing, has developed software that records data use, phone use, and battery-charge levels. The software is designed to run on devices that use Windows Mobile or the Android operating system. The Android version can also track the data sent and received by individual applications.
“One major problem we all experience with smart phones is that the batteries don’t last long enough,” says Falaki, who will present a paper next month at the Internet Measurement Conference in Melbourne, Australia, on more than 2,000 days of data collected from eight Windows Mobile and 35 Android users. “By studying how people use [the phones], we can find ways to match devices and networks to people.”
For example, the tracking application uncovered data suggesting that a tweak to the hardware of two phones made by the Taiwanese manufacturer HTC– could save approximately 40 percent of the power consumed by their radios. These handsets automatically switch off the radio after being idle for 17 seconds, a tactic used by all handsets and often with a similar timeout value. But that is a poor match with the very “bursty” way that smart-phone users access data, says Falaki. “People take the phone out of their pocket, interact with it for a few minutes, and then don’t use it for a relatively long time after,” he says.
Logs of data use showed that after a burst of activity, users rarely needed more data in the subsequent 17 seconds, so the radio was often left on needlessly. In fact, some 95 percent of data packets were sent or received within 4.5 seconds of the last one. Resetting the device so that the radio powered down after 4.5 seconds would consume 40 percent less power without affecting performance, says Falaki.
“These ‘tail times’ are larger than they need to be,” says Arun Venkataramani, an assistant professor at University of Massachusetts, Amherst who studies power use in mobile devices. “From an application and user perspective, there’s significant room for improvement.” The Microsoft-UCLA data agrees with results from his own experiments looking at the energy costs of cell-phone timeout periods, he says.
However, the length of time a cell phone waits before sleeping is partly set by cell phone networks that have other concerns than a user’s battery life, Venkataramani adds. For example, tail times are sometimes made long to reduce the work of cell towers, which must exchange control messages with devices whenever they wake or sleep. Gathering more data on user behavior may help raise the profile of the battery-life problems, says Venkataramani, but carriers are still likely to want to put their networks first.
Falaki says devices, apps, and wireless networks should be designed with more regard to how people actually use their devices, but that because smartphones are still relatively new, little is known about that. Previous studies have used data about wireless networks as a whole instead of individual user behavior, or lab-based experiments where phones transfer test data and patterns of everyday use are not encountered, he says. “In contrast, we capture real usage patterns as people used their devices normally.”
Lin Zhong, who leads the Efficient Computing Group at Rice University, points out that very little is really known about how people use smart phones. “The UCLA-Microsoft data set is small, but they have already highlighted some interesting results,” he says.
Zhong and his Rice colleagues developed tracking software that runs on hacked, or jailbroken, iPhones and can capture network data, battery information, and the times when people use specific applications or features. His research group is in the final four months of a yearlong trial using the software to capture every aspect of the way 35 people use their iPhone 3GS devices.
“It sounds like a virus in the device, but we do it in a way that preserves privacy and does not interfere with the person when they are using their phone,” says Zhong. For example, although records are logged of all phone calls, the numbers called are obfuscated.
The Rice group is using its data to create a new kind of cell-phone coverage map: one created by collecting the actual experiences of real users. “If you look at the coverage map for, say, AT&T, it will tell you there is full 3G coverage in Houston, but we know that’s not what you experience,” says Zhong. “We can collect a very fine-grained network map.”
This is an example of how cell-phone-logging apps can be of use to more than just researchers, he points out. Future phone users could benefit from an app that logs and shares the network strength experienced by their phone to build a dynamic map of cellular network coverage and Wi-Fi hot spots, he says.
Become an MIT Technology Review Insider for in-depth analysis and unparalleled perspective.Subscribe today