云计算系统性能保障若干关键技术研究

Because of the massive scales and complicated interactions, it is difficult to cloud computing systems to predict all online situations during the phrase of development. How to guarantee the performance of cloud computing systems in real time has become one of the core technical challenges in the related international areas. This project tries to ensure the performance of cloud computing systems by ways of the system behavior monitoring and performance anomalies diagnosing. In views of the new challenges, the monitor-supporting mechanisms will be studied in aspect of the runtime behavior monitoring; while the approaches of high-precision detecting, fine-granularity localizing and high-efficiency fixing will be researched in term of performance anomalies diagnosing. We research four kinds of key approaches, which involves massive requests oriented runtime behavior monitor-supporting mechanism, request execution paths based performance anomalies detecting approach, multiple replicated instances based performance anomalies localizing approach and feature matching based performance anomalies fixing approach. This project targets to construct a set of theories and mechanisms for the performance guarantee of cloud computing systems and promote the research level as well as the independent innovational ability of our country in the related areas.

云计算系统因其规模巨大、交互复杂等因素在开发阶段难以预料线上环境面临的所有情况，其服务性能难以实时满足用户的体验需求。如何对云计算系统的性能进行实时保障已成为目前国际相关领域面临的核心技术挑战。本项目力图从运行时行为监测与性能异常诊断两个方面保障云计算系统的性能。针对这两个方面面临的新挑战，本项目在运行时行为监测方面，将着重研究海量请求条件下的监测支撑机制；在性能异常诊断方面，将着重研究高精度的检测方法、细粒度的定位方法和高效率的修复方法。具体研究包括面向海量请求的云计算系统运行时行为监测支撑机制、基于请求执行路径的云计算系统性能异常检测方法、面向多副本实例的云计算系统性能异常定位方法和基于特征匹配的云计算系统性能异常修复方法四个方面。本项目旨在建立一套切实有效的保障云计算系统性能的理论与机制，提升我国在相关领域的研究水平和自主创新能力。

随着云计算系统在各领域广泛的应用，如何对云计算系统的性能进行实时保障已成为学术界和工业界面临的核心技术挑战。课题以云计算系统的性能维护为背景，重点围绕性能监测机制、性能异常诊断和性能优化三个方面展开研究，形成了较为系统、深入和具有原创性的技术成果体系。.在性能监测机制方面，设计了基于用户请求路径的分布式系统监测平台，并基于此平台收集了一个细粒度、多场景的数据集；在性能异常诊断方面，分别提出了基于运行时验证的性能异常检测方法、基于路径分割的性能异常诊断方法和基于时间自动机的性能异常诊断方法；在性能优化方面，设计了面向容器云平台的镜像分发性能优化机制和存储敏感的云服务优化重部署机制。课题相关研究成果成功应用于南部战区某指控系统底层云平台和阿里云云数据服务（RDS）生产系统，取得了显著效果。.课题在Transactions on Service Computing等高水平期刊和会议共计发表学术论文10篇，其中SCI检索2篇、EI检索8篇。培养博士生2名，硕士生3名。