Mpower : on how to effectively predict the time to live for mobile devices

Mobile devices are more and more popular, due to the growing number of functionalities they provide. Nonetheless, they have to deal with reduced computational power and, first of all, limited battery life. Within this scenario, it is important to develop technologies to save as much energy as possible, in order to increase the battery Time-To-Live (TTL). Anyhow, the problem is not just to save battery life "a priori": the device has to allow the user to achieve his goals, e.g., make a phone call in two hours when the battery level is at 15 percent. In order to suggest to the user the right policy to save battery life w.r.t. his needs, the device TTL must be predicted: a practical way of doing this is to model the battery discharging curve, since TTL coincides with the time the battery level reaches zero level. A power modeling methodology for mobile devices is then necessary considering also that the batteries TTL is highly influenced by external factors like the user usage profile or 3G signal strength. This thesis propose MPower, a system able to accurately predict the TTL of Android devices in their real usage scenario. It is mainly composed of a mobile app and a remote server. On the one side, the mobile app is in charge of retrieving information about the device in its real usage context, without influencing its discharge behavior. On the other side, the remote server is meant to receive, store and process the gathered data, to analyze the device power consumption: no computation is then done on the device. Once the power model has been computed on the server, the power model is sent back to the device: it consists of a list of configurations (i.e., states of the device hardware components, such as Wi-Fi, Bluetooth, 3G module, etc.) and the predicted TTL for every one of them. The user can then set the device in a particular configuration in order to achieve the related TTL. Statistical tests point out the precision produced by MPower: usual values has a Mean- Hourly-Error (MHE) near 0, with standard deviations below 4. Moreover experimental results showed that, the MPower application does not significantly impact on the device power consumption.