CSR: Small: Power-Efficient Multicore Scheduling for Network Applications

CSR：小型：网络应用的高能效多核调度

• 批准号: 1216014
• 负责人: Laxmi Bhuyan
• 金额: $ 40万
• 依托单位: University of California-Riverside
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2012
• 资助国家: 美国
• 起止时间: 2012-10-01 至 2017-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1216014&HistoricalAwards=false
• 关键词: CSR; Small; Power; Efficient; Multicore

High-speed multicore computers are widely available at low cost. They have the potential to process high speed network applications, such as deep packet inspection, and multi-channel real-time streaming of high-definition audio and video. However, they need better scheduling approaches to address the demand of high throughput and low latency of these applications while reducing energy consumption. Current approaches do not consider real-time demands, network traffic patterns and details of the specific multicore architectures. Lack of power-aware core scheduling and thermal management has also led to a great challenge in energy efficiency. Reducing power consumption saves energy cost, reduces the need for building additional cooling, and improves computer hardware reliability.This project makes fundamental contributions to the design and applications of high-demand network systems. It provides packet level scheduling solutions that improve throughput, reduce latency and energy consumption. The project develops and incorporates network traffic models, architectural details, power consumption models, and thermal models simultaneously for coordinated workload processing and energy consumption. Quality-of-service (QoS) of packet streams/connections is also considered by improving the jitter and out-of-order packet departures. Superiority of the proposed technical approaches is verified through real implementation of network applications and performance and power measurements on commercial multicore architectures. An open-source framework is developed for multicore scheduling, and is made available to the researchers along with optimized multithreaded codes for network applications.

高速多核计算机以低成本广泛可用。它们具有处理高速网络应用的潜力，例如深度数据包检测以及高清音频和视频的多通道实时流。然而，他们需要更好的调度方法来解决这些应用程序的高吞吐量和低延迟的需求，同时降低能耗。目前的方法不考虑实时需求，网络流量模式和具体的多核架构的细节。缺乏功耗感知的核心调度和热管理也导致了能源效率方面的巨大挑战。降低功耗可以节省能源成本，减少对建筑物额外冷却的需求，并提高计算机硬件的可靠性。该项目为高需求网络系统的设计和应用做出了基础性贡献。它提供数据包级调度解决方案，可提高吞吐量、减少延迟和能耗。该项目同时开发并整合了网络流量模型、架构细节、功耗模型和热模型，以协调工作负载处理和能耗。还考虑通过改善抖动和无序分组离开来提高分组流/连接的服务质量（QoS）。通过真实的网络应用实现以及商用多核架构上的性能和功耗测量，验证了所提出的技术方法的可行性。一个开源的框架开发的多核调度，并提供给研究人员沿着优化的多线程代码的网络应用程序。