Energy locality: processing/communication/interface tradeoffs to optimize energy in mobile systems

Abstract

Summary form only given, as follows. As computers continue to shrink, research and commercial interest in mobile/wearable computing is rapidly growing. Unlike traditional desktop computing in which the user is required to come to the computer, mobile computing brings the computer to the user. Mobile/wearable computers represent the next evolutionary step in the trend toward more people-centric computing. One of the key problems with mobile/wearable computing is energy consumption. Battery weight for mobile/wearable computers often exceeds the weight of all other components combined. In order to make mobile/wearable computing widely applicable, major advances in reducing power consumption and battery weight are needed. While a "Moore's Law" exists for power consumption of microprocessors with mW/MIPS decreasing by a factor of ten every five year, there is no such similar trend in wireless communications. This suggests that future wearable computers will be communications bound. In fact, we estimate that nearly 80% of the power consumed by wearable computers can be due to communications. Trading off energy expensive communication for energy cheap computation through effective partitioning of control and data can result in significant energy savings. Examples and measurements will illustrate how the use of proxies can reduce power consumption due to communications by several orders of magnitude. In addition, the interface design must be carefully matched with user tasks and balanced against energy consumption. Many complex and interrelated issues determine the balance between ease-of-use and power consumption. Simply trading off ease-of-use for lower per operation power consumption may result in higher task energy consumption due to the increase in the number of operations needed to traverse a less intuitive interface. The effect of user interface on energy consumption can be evaluated by developing several different interfaces and measuring and comparing the ease-of-use and energy consumption. In conclusion, an architecture that supports these studies will be introduced. The Spot wearable computer includes a dozen power monitors that can be read under software control to determine which subsystems are active and their power consumption during an application.