The popular IEEE 802.11 standard defines a power saving mode that keeps the network interface in a low power sleep state and periodically powers it up to synchronize with the base station. The length of the sleep interval, the so called beacon period, affects two dimensions, namely application performance and energy consumption. The main disadvantage of this power saving policy lies in its static nature: a short beacon period wastes energy due to frequent activations of the interface while a long beacon period can cause diminished application responsiveness and performance. While the first aspect, reduction of power consumption, has been studied extensively, the implications on application performance have received only little attention. We argue that the tolerable reduction of performance or quality depends on the application and the user. As an example, a beacon period of only 100 ms slows down RPC-based operations like NFS dramatically, while the user will probably not recognize the additional delay when using a web browser. If at all, known power management algorithms guarantee a system wide limit on performance degradation without differentiating between different application profiles. This work presents an approach to identify on-line the currently running application class by a mapping of network traffic characteristics to a predefined set of application profiles. We propose a power saving policy which dynamically adapts to the detected application profile, thus identifying the application- and user-specific power/performance trade-off. An implementation of the characterization algorithm is presented and evaluated running several typical applications for mobile devices.
展开▼
