Collaborative Research: Algorithmic Support for Power Aware Computing and Communication

协作研究：功耗感知计算和通信的算法支持

• 批准号: 0514058
• 负责人: Kirk Pruhs
• 金额: $ 15万
• 依托单位: University of Pittsburgh
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-07-15 至 2009-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0514058&HistoricalAwards=false
• 关键词: Collaborative; Research; Algorithmic; Support; Power

Intellectual Merit: The power consumption rate of computing devices has been increasing exponentially. This makes it increasingly difficult to supply energy to devices and to cool these devices. Poweraware computation is especially important in the domain of sensor networks which are composed of small battery-powered nodes. Power management in sensor networks is viewed as so critical that it must be dealt with at all layers of the protocol stack.Many power management techniques have been proposed and implemented. Most of these techniques are similar in that they reduce or eliminate power to some or all components of the device. However, there is an inherent conflict between power management and performance; in general, the more power that is available, the better the performance that can be achieved. As a result, it is generally proposed that power reduction techniques be preferentially applied during times when performance is lesscritical. However, this requires a policy to determine how essential performance is at any given time and how to apply a particular power reduction technique. For example, to use the frequency scaling technique, where the speed of the clock is changed dynamically, one needs a policy to set the speed at each point in time. There is a growing consensus that these policies must incorporate information provided by applications and high levels of the operating system, and that current tools and mechanisms for power management are inadequate and require more research. The authors propose to formalize powermanagement problems as optimization problems, and then develop algorithms that are optimal by these criteria. The goal of this research is to develop effective algorithms for specific problems within the domain of power management, as well as to build a toolkit of widely applicable algorithmic methods for problems that arise in energy-bounded and temperature-bounded computation. The authors propose to initially focus on problems that deal with speed scaling and power-down techniques, since these are currently the dominant techniques in practice.Broader Impacts: The authors propose to both develop fundamental theoretical techniques, and to apply these techniques to attack timely and important applications in computer systems. Both PIs have an established track record of working closely with researchers in applied areas to ensure that the theoretical models developed match the associated real-world problems. This is essential for theoretical results to have an impact. This work will continue to foster this very productive cross-fertilization betweenthese experimental systems researchers and theoretical computer science. The students supported under this grant will be influenced by this philosophy of research. They will be trained to be proactive in working with researchers in applied domains to bring important and interesting problems into the theory community. They will also be encouraged to publish the resulting work in systems as well as theory conferences to ensure that new algorithmic discoveries have an impact. As part of this project, the authors also plan to continue outreach work to high schools students, encouraging underrepresented groups to choose careers in technology related fields. They have developed a talk outlining diverse opportunities within computer science and plan to involve graduate and undergraduate students in presenting this talk at local high schools. The work in this proposal is featured in the talk. In addition, they plan to involve students from underrepresented groups in research projects related to power management.

智力优势：计算设备的功耗呈指数级增长。这使得向设备提供能量以及冷却这些设备变得越来越困难。节能计算在由小型电池供电节点组成的传感器网络领域尤其重要。传感器网络中的电源管理被认为非常重要，必须在协议栈的所有层上进行处理。已经提出并实现了许多电源管理技术。大多数这些技术的相似之处在于它们减少或消除了设备的部分或所有组件的功率。然而，电源管理和性能之间存在固有的冲突；一般来说，可用功率越多，可以实现的性能就越好。因此，通常建议在性能不太重要的时候优先应用功率降低技术。然而，这需要一种策略来确定性能在任何给定时间的重要性以及如何应用特定的功耗降低技术。例如，要使用时钟速度动态改变的频率缩放技术，需要一种策略来设置每个时间点的速度。人们越来越一致地认为，这些策略必须纳入应用程序和操作系统高层提供的信息，并且当前的电源管理工具和机制还不够充分，需要更多的研究。作者建议将电源管理问题形式化为优化问题，然后开发根据这些标准优化的算法。本研究的目标是针对电源管理领域的特定问题开发有效的算法，并针对能量有限和温度有限计算中出现的问题构建广泛适用的算法方法工具包。作者建议首先关注处理速度缩放和断电技术的问题，因为这些是目前实践中的主导技术。更广泛的影响：作者建议开发基础理论技术，并应用这些技术进行及时攻击以及计算机系统中的重要应用。两位 PI 都拥有与应用领域的研究人员密切合作的良好记录，以确保开发的理论模型与相关的现实问题相匹配。这对于理论结果产生影响至关重要。这项工作将继续促进这些实验系统研究人员和理论计算机科学之间非常富有成效的交叉融合。接受这项资助的学生将受到这种研究哲学的影响。他们将接受培训，积极主动地与应用领域的研究人员合作，将重要且有趣的问题带入理论界。还将鼓励他们在系统和理论会议上发表成果，以确保新的算法发现产生影响。作为该项目的一部分，作者还计划继续向高中生开展外展工作，鼓励代表性不足的群体选择技术相关领域的职业。他们制定了一场演讲，概述了计算机科学领域的各种机会，并计划让研究生和本科生在当地高中进行演讲。演讲中重点介绍了该提案中的工作。此外，他们计划让代表性不足群体的学生参与与电源管理相关的研究项目。