On Energy, Adaptation, and the Death of Frames
msra(2006)
摘要
Increases in computing capabilities of mobile devices have led to the use of multimedia applications that have high pro- cessor and bandwidth resource requirements, each of which consume significant energy. However, battery capacities have not kept pace, driving the design of energy-aware applica- tions. Traditionally multimedia applications have used en- coding techniques to deal with bandwidth constraints, es- sentially trading off CPU for networking resources. How- ever, in the context of energy constraints, it is not clear that this CPU-intensive approach is the most efficient. Making correct trade off decisions requires information about costs of both the CPU and the network. In addition to energy- savings by adaptive applications, further energy conserva- tion can be achieved by leveraging application layer informa- tion at the network layer. Because of their ability to tolerate loss, multimedia applications present unique opportunities to the design of such energy-efficient network protocols, de- spite their strict timing constraints. Essentially, transport protocols can use application level information to make in- telligent decisions about when to perform frame recovery, positively impacting system energy conservation. Previous adaptive systems concentrate on sharing resource informa- tion between the application and the network in only one direction and so perform sub-optimally. Therefore, it is nec- essary to design cooperative solutions that share resource information in both directions. To this end, we present a data-oriented energy model that exposes cross-layer inter- actions and enables specific optimizations in the applica- tion and network layers. We demonstrate that by passing a minimal amount of information in both directions, an adap- tive application paired with our adaptive transport proto- col, Reaper, can achieve significant energy savings, which we verify through a system implementation.
更多查看译文
关键词
mobile computing,computer science
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络