Context for This Work T System Support for Mobile, Adaptive Applications

prospect of anytime, anywhere network access presents both opportunity and challenge to application writers. Providing connectivity to those on the move enables entirely new services while expanding the reach of current applications. Unfortunately, taking advantage of pervasive networking is difficult. The quality of network con-nectivity afforded mobile users is extremely turbulent: it changes frequently, dramatically, and without warning. Applications must somehow cope with these changes. There are many reasons for this dynamic behavior. An individual wireless channel is subject to path loss, fading, and environmental interference [1]. Together, these can have a significant impact on the performance of the channel. Further, overlay networks arrange wireless coverage as a set of overlapping technologies, each providing a different tradeoff between bandwidth, coverage, cost, and reliability [2]. How can applications deal with such variation? Ideally, they would take advantage of high-quality connectivity when available, but behave reasonably over networks with poor performance. In the past, systems have dealt with variation by trading plentiful resources for those that are scarce. For example, Web caches [3] attempt to insulate clients from the vagaries of wide-area network performance by spending disk space. Unfortunately, such techniques are not able to hide the orders-of-magnitude changes in performance that are all too common in wireless networks. Rather than rely on the system to manage resources transparently , applications must themselves adapt to prevailing network characteristics. This article presents an overview of our experience with Odyssey, a platform for adaptive mobile data access. We have developed several applications for Odyssey, including a Web browser, a video player, and a speech recognition system. Each of these applications is described in more detail in a forthcoming paper [4]. Odyssey's approach to adaptation is to adjust the quality of accessed data to match available resources. For example, when faced with a sharp decrease in bandwidth, a Web brows-er might ask for more highly-compressed images; a video player may reduce frame rate or frame quality of the stream; and a map viewer may filter out small or irrelevant features. In order to trade data quality for resource consumption and performance , one must first have a notion of quality. We introduce fidelity, Odyssey's notion of data quality, and present a few simple examples of its use. We present Odyssey's division of labor between applications and the operating system for making adaptive decisions. In this division, the system provides the mechanisms enabling adaptation, leaving applications …