Battery Properties Battery-driven Dynamic Power Management

0740-7475/01/$10.00 © 2001 IEEE March–April 2001 THE ACTIVITY OF SEVERAL COMPONENTS in a computing system is event-driven. For example, the activity of display servers, communication interfaces, and user interface functions is triggered by external events, and it is often interleaved with long, idle periods. An intuitive way to reduce average power dissipated by the whole system consists of shutting down resources during periods of inactivity. In other words, one can adopt a dynamic power management (DPM) policy that dictates how and when various components should be shut down according to a system’s workload. Workload-driven DPM can be very effective, thanks to sophisticated policies, based on complex computational models (such as Markov chains) proposed in the recent literature.1 We observe, however, that minimum average power is not always the objective when designing battery-operated, mobile applications. Rather, what really matters for this kind of system is ensuring long battery lifetime. Average power reduction and battery lifetime extension may be numerically far apart.2 This implies that optimizations for minimum average power may not be equally effective in extending battery lifetime, and vice versa. Our work moves from the assumption that taking battery’s charge state into account while managing the system helps in maximizing the time of operation of portable devices. We describe several DPM policies specifically tailored to battery lifetime maximization. In particular, we introduce a class of closed-loop policies, whose decision rules used to control the system operation state are based on the observation of a battery’s output voltage (which is related, nonlinearly, with the charge state). This is in contrast with open-loop solutions that reach decisions about component shutdown independently from battery voltage measurement. Open-loop policies are normally simpler, but less effective, than closed-loop ones; they represent a viable option when cost constraints prevent the use of a voltage sensor on the battery terminals. On the other hand, the distinguishing feature of closed-loop policies is that they control system operation based on the observation of both system workload and battery output voltage. As a consequence, they can dynamically adapt a component’s shutdown scheme to the actual battery charge state.