云服务器性能衰退预测与多级再生模型研究

The rapid development of cloud computing technology brings the requirements on applications in cloud computing systems into an extend area. The need for reliability, availability and performance has increased in cloud computing systems. Multi-tenancy enables the sharing of resources and costs across a large pool of users. However, the complex interactions between cloud service and multi tenant application lead the cloud computing environment into the dynamic runtime scenarios, in which the long-running server system shows gradual resource consumption, reduced quality of service or even unavailable services. When the performance degradation appears, software rejuvenation could be taken timely to restore system to the healthy state, thus improving the performance. The project is to research the problem of performance degradation and the preventive maintenance method of software rejuvenation for the server system in the cloud computing environment, based on the characteristics of multidimensional performance parameters and complex runtime environment. Firstly, through monitoring the runtime states and the related performance parameters, the formal method is to be researched to analyze the relationship between performance parameters, extract the system key parameters, and identify the degradation process in cloud server system. Secondly, the research on the performance degradation prediction model and the quantification analysis method, are to be proposed to analyze the system performance decline trend and calculate the performance degradation threshold, if it is violated, the system crashes. Finally, based on the performance degradation prediction model, the multi-stages and hierarchical software rejuvenation modeling policy under minimum maintenance cost and maximum availability is to be researched, and the healthy rejuvenation mechanism is to be realized. The simulation and prototype will be put forward to evaluate and demonstrate the research results. The project will improve the efficiency and quality of system service, reduce difficulty and cost of online system maintenance, and ensure the system reliability and availability.

随着云计算技术在计算机系统中的广泛应用，云用户对服务可用性和系统可靠性的要求与日俱增。云服务器系统在长期运行过程中，由于云服务和多租户应用层的复杂交互，以及系统运行环境的动态变化，导致系统性能下降，服务质量降低甚至不可用。及时采取必要的软件再生措施，可有效地维护系统运行过程，提高系统性能。本项目研究云服务器系统的性能衰退预测及软件再生问题。针对云服务器性能参数多且运行环境复杂的特点，监测当前运行的云服务器性能参数信息，分析系统性能衰退过程中各参数间的关联关系，筛选系统关键性能参数，识别系统性能衰退过程；研究系统性能衰退预测建模方法，预测系统性能衰退趋势，计算性能衰退阈值，为软件再生手段提供依据；基于系统的分层结构特征和多粒度运行状态，建立多级优化再生模型和分层健康再生机制，并搭建原型系统验证研究成果的有效性。本研究可用于提高云服务器系统的服务质量与与效率，降低系统在线维护难度与成本。

云服务器系统出现性能衰退，服务质量降低甚至不可用，影响系统的可靠性和可用性，特别对安全关键系统，将带来无可估量的经济损失。采取软件再生对云服务器系统进行有效地维护，可以提高系统的服务性能。本课题针对云服务器系统在多重负载发生场景下，研究了系统性能衰退的参数特征提取、识别和预测问题，负载任务调度和资源分配问题，分层优化再生问题及可靠性评估问题。课题组共发表论文11篇，其中SCI检索论文3篇，EI检索论文7篇，申请国家发明专利4项，获得软件著作权2项，培养研究生7名，在国际会议上做口头报告3次。.本课题主要成果包括：(1)监测并提取了云服务器系统关键性能指标，构建了基于时间序列的系统性能衰退预测模型，识别云系统性能衰退过程；(2)为了分析负载任务调度策略及传输性能对云系统运行状态和可靠性的影响，建立最低消耗成本的任务调度优化模型，并设计了面向云计算的数据中心网络传输协议；(3)通过分析云系统的应用服务模式，提出了面向租户服务响应时间的资源消耗度量模型，并给出了基于成本优化的多租户优化放置策略；(4)基于性能衰退模型，提出了一种马尔可夫随机再生模型，并设计了基于几何序列递减随机模型的云系统再生控制策略；(5)基于云计算系统的分层体系结构特征，提出了动态故障树的再生建模及优化方法；(6)实现实验原型系统，对所提模型、理论及方法进行了验证。.本课题这些理论研究成果将为云计算环境下服务器系统的可靠运行提供理论基础，为云服务的在线维护策略提供技术支撑。