CAREER: Decentralized Control and Optimization Techniques for Autonomic Performance Management of Distributed Computing Systems

职业：分布式计算系统自主性能管理的分散控制和优化技术为了在不确定和动态的环境中有效地运行分布式计算系统，必须动态调整多个与性能相关的参数，如应用程序的资源供应、应用程序之间的相对优先级和应用程序运行模式，以匹配随时间变化的应用程序需求和运行条件。为了应付它们日益增长的规模和复杂性，这些系统必须变得自治，也就是说，能够根据管理员提供的高级目标进行自我管理。本项目旨在发展设计自主计算系统的理论与实践。具体的研究目标是：(1)开发一个分散的框架，使用模型预测和最优控制的概念来管理大规模计算系统；(2)从稳定性和鲁棒性两个方面分析控制算法的性能；(3)在Blade-server集群上使用基于跟踪的仿真和实际实现验证控制技术。该项目的预期结果是一组技术，可以使用经过验证的控制理论技术将大量系统管理任务转换为系统过程。这将降低系统运行成本，同时确保支持业务、科学和工程社区的应用程序的顺利运行。获得的结果将通过相关的期刊和会议传播到研究界，软件工件将通过项目网站免费提供。该教育计划旨在：(1)作为宾夕法尼亚州切斯特社区教育支持项目的一部分，指导高中生学习计算机科学；(2)加强德雷塞尔大学的本科课程设置；(3)开设一门新的自主计算研究生课程。

CAREER: Decentralized Control and Optimization Techniques for Autonomic Performance Management of Distributed Computing SystemsPI: Nagarajan Kandasamy, Drexel University, Philadelphia, PA 19104SummaryTo operate distributed computing systems effectively in an uncertain and dynamic environment, multiple performance-related parameters such as resource provisioning to applications, relative priority between applications, and application operational modes, must be dynamically tuned to match time-varying application requirements and operating conditions. To cope with their growing scale and complexity, these systems must become autonomic, i.e., capable of managing themselves given high-level objectives from administrators. This project aims to develop the theory and practice of designing autonomic computing systems. The specific research objectives are to: (1) Develop a decentralized framework, using concepts from model-predictive and optimal control, to manage large-scale computing systems; (2) Analyze the performance of the control algorithms in terms of stability and robustness properties; and (3) Validate the control techniques using trace-based simulations and actual implementations on a Blade-server cluster. The expected outcome of this project is a set of technologies to convert a significant number of system management tasks into systematic processes using proven control-theoretic techniques. This will reduce system operating costs while ensuring the smooth operation of applications supporting the business, science, and engineering communities. The obtained results will be disseminated to the research community via relevant journals and conferences, and software artifacts will be freely available via the project web site.The educational plan seeks to: (1) Tutor high-school students in the computer sciences as part of a community-based educational support project in Chester, Pennsylvania; (2) Enhance undergraduate course offerings at Drexel University; and (3) Create a new graduate course on autonomic computing.