DCS: Retrofitting a Flow-oriented Paradigm in Commodity Operating Systems for High-Performance Computing

DCS：在商品操作系统中改造面向流的范式以实现高性能计算

• 批准号: 0541093
• 负责人: Angelos Keromytis
• 金额: --
• 依托单位: Columbia University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2006
• 资助国家: 美国
• 起止时间: 2006-01-01 至 2009-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0541093&HistoricalAwards=false
• 关键词: DCS; Retrofitting; Flow; oriented; Paradigm

BackgroundThe PI proposes an operating system and application architecture that removes the memory and CPU from the data path for applications that handle high-bandwidth data flows. The role of the operating system becomes that of data-flow management, while applications are concerned purely with signaling. This design parallels the evolution of modern network routers and has the potential to enable high-performance I/O for end systems, as well as fully exploit recent trends toward programmable peripheral (I/O) devices. An alternative view is that hardware devices are composed into virtual processing pipelines, completely removing the CPU and main memory from the data-intensive tasks for which they are increasingly unsuitable due to performance considerations. The Pis hypothesis is that application-specific data-flow handling policies can be executed inside the operating system kernel, which composes and configures peripherals, manages flows, and handles exceptions. Such functionality can be retrofitted to existing operating systems and applications, instead of requiring a complete redesign. Several interesting research issues with respect to safety, flow scheduling, error handling, and resource management arise in the proposed architecture.Intellectual MeritThe proposed architecture is a first step toward abandoning the concept of memorycentriccomputing, a paradigm ingrained in computer scientists for several decades. This shift in architecturalthinking is both necessary for performance reasons and natural, given the increasing use of task-specificand programmable hardware. The proof-of-concept prototype indicates that the proposed system can at least double I/O throughput for large flows, and is orthogonal to other performance optimization approaches (e.g., faster interconnection buses). This research intends to bring together such diverse fields ofComputer Science as networking, operating systems, embedded system design, programming languages, andsecurity, and to change the way these are taught, from individual subjects to interdependent items that require (and promote) wide knowledge and creative thinking. Broader Impact-This architecture will enable high-performance applications and allow more efficient use of server resources-It will also allow experimentation and deployment of new types of system architectures, allowing use of novel hardware designs, device interconnection strategies, and load-balancing and scheduling algorithms.-This approach will also eliminate one commonly perceived problem with the wide-spread use of secure protocols and cryptography, that of performance.-The PI hopes to enable a new class of systems that are able to support highperformance applications across a wide range of areas of computer science -Finally, by fully taking advantage of computing systems capabilities, users can reap considerable savings in equipment spending.-The PI plans to involve creating cross-subject projects and an undergraduate laboratory class that exposes students to interesting projects and research at the intersection of systems, networks, and security. Many of these projects will allow students to cross the hardware/software boundary, which few current professionals can do despite increasing demand

背景PI 提出了一种操作系统和应用程序架构，该架构将内存和CPU 从处理高带宽数据流的应用程序的数据路径中移除。操作系统的角色变成了数据流管理，而应用程序只关心信号发送。该设计与现代网络路由器的发展并行，具有为终端系统实现高性能 I/O 的潜力，并充分利用可编程外围 (I/O) 设备的最新趋势。另一种观点是，硬件设备被组成虚拟处理管道，将CPU和主内存从数据密集型任务中完全移除，而出于性能考虑，它们越来越不适合这些任务。 Pis 假设是，特定于应用程序的数据流处理策略可以在操作系统内核内部执行，该内核组成和配置外设、管理流并处理异常。此类功能可以针对现有操作系统和应用程序进行改造，而不需要完全重新设计。所提出的体系结构中出现了有关安全、流调度、错误处理和资源管理的几个有趣的研究问题。 智力优点所提出的体系结构是放弃以内存为中心的计算概念的第一步，这一概念在计算机科学家中根深蒂固了几十年。鉴于任务特定和可编程硬件的使用越来越多，这种架构思维的转变既是出于性能原因所必需的，也是自然的。概念验证原型表明，所提出的系统至少可以使大流量的 I/O 吞吐量增加一倍，并且与其他性能优化方法（例如更快的互连总线）正交。这项研究旨在将计算机科学的不同领域结合在一起，例如网络、操作系统、嵌入式系统设计、编程语言和安全性，并改变这些领域的教学方式，从单独的科目到需要（和促进）广泛知识和知识的相互依赖的项目。创造性思维。更广泛的影响——该架构将实现高性能应用程序，并允许更有效地利用服务器资源——它还将允许试验和部署新型系统架构，允许使用新颖的硬件设计、设备互连策略以及负载平衡和调度算法。-这种方法还将消除广泛使用安全协议和加密技术时常见的一个问题，即性能问题。-PI 希望启用一类新的系统，能够支持广泛的高性能应用程序计算机科学领域的范围-最后，通过充分利用计算系统功能，用户可以节省大量设备支出。-PI 计划创建跨学科项目和本科生实验室课程，让学生接触系统交叉点的有趣项目和研究、网络和安全。其中许多项目将允许学生跨越硬件/软件边界，尽管需求不断增加，但目前很少有专业人士能够做到这一点