CSR-PDOS: Resource Allocation Optimization for Quantitative Slowdown Differentiation in Multi-tier Server Clusters

互联网服务已经成为推动可扩展和质量感知的分布式计算技术的重要应用类别。服务差异化就是通过有偏见地将资源分配给分类的用户请求，从而提供不同的质量水平，以满足互联网服务的需求。在这项研究项目中，研究人员采用分析和组织的方法来检查流行的多层服务器集群中用于量化服务区分的资源管理技术。多层集群提供了新的计算机会，但也带来了诸如跨层性能依赖和并发限制等挑战。该项目设计了合理的差异化模型，既考虑了全系统范围的资源分配优化，也考虑了用户侧公平的服务质量提供。它从一个新的重要度量--减速--提供服务差异化和系统性能分析，减速定义为作业延迟与其服务时间的比率。这些成果是在排队论和控制论的基础上，为多层服务器集群上可预测和可控的服务区分和性能隔离提供创新的资源分配方法。研究的意义在于提供具有质量意识的互联网计算技术。其影响在于从独特的减速角度更新对计算机系统性能的理解，以及为提高分布式系统的可扩展性和可靠性而进行的质量控制。潜在的应用包括一种对抗快闪人群流量的艺术品。研究结果将以出版的技术报告的形式向公众发布。该项目将帮助该协会为流行的互联网服务开发具有质量意识的应用程序和可销售的计算技术。

Internet services have become an important class of driving applications for scalable and quality aware distributed computing technologies. Service differentiation is to provide different quality levels to satisfy requirements of Internet services by allocating resources to classified user requests in a biased way. In this research project, the investigators take an analytical and organized approach to examine resource management techniques for quantitative service differentiation in popular multi-tier server clusters. Multi-tier clusters provide new computing opportunities but also challenges such as cross-tier performance dependencies and concurrency limits. The project designs sound differentiation models that consider both system-wide resource allocation optimization and user-side fair service quality provisioning. It provides service differentiation and system performance analysis from a new important metric, slowdown, defined as the ratio of a job's delay to its service time. The deliverables are innovative resource allocation approaches for predictable and controllable service differentiation and performance isolation on multi-tier server clusters, based on the foundations of queuing theory and control theory. The significance of the research is to provide quality aware Internet computing techniques. The impacts lie in renewing the understandings of computer system performance from the unique slowdown perspective and in quality control for scalability and dependability enhancements of the distributed systems. Potential applications include an artifact against flash-crowds traffic. The research results will be disseminated to the public as published technical reports. The project will help the society develop quality aware applications and salable computing technologies for popular Internet services.