A dominant factor in estimating the cost of computing systems is the component of ownership cost associated with energy consumption. Tasks such as keeping data regularly updated involve an energy cost and we will now have to design systems that explicitly expose different tradeoffs to achieve energy savings.

I will describe some ideas that were motivated by work with sensor networks to illustrate how we can save energy by tolerating some staleness in the data we seek. In particular, I will describe some methods for scheduling data updates from sensor networks by exploiting some understanding of (i) the stochastic process affecting of data changes and (ii) the distribution governing the arrival of user queries. Such ideas are also applicable to large multiprocessing systems where one may want to relax data consistency requirements to reduce overheads and save energy. If time permits, I will broaden the scope of the discussion to include tradeoffs between quality of service and energy consumption in computer systems by considering the effect of the imprecise computation model.