面向高干扰共享云的性能异常检测方法研究

Cloud computing uses virtual machine consolidation technology to improve the utilization of physical resources, but at the same time brings the problem of shared resources competition and interference, leading to application performance anomalies. It is meaningful to effectively detect and accurately locate the performance anomalies in cloud computing systems, making it better adapt to the characteristics of large-scale, high-complexity, resource-sharing of cloud computing systems. This project aims to solve the problems such as poor real-time performance, poor detection accuracy and high cost of the current detection methods for cloud computing systems. We study the performance anomaly issues from two aspects: anomaly detection based on anomaly feature extraction and cross-level correlation analysis; anomaly source localization based on application request path tracking and dynamic dependency graph construction. We will finally build a prototype of performance anomaly diagnosis for highly interference shared cloud systems. The results of the research will be able to improve the reliability of cloud systems, enhance cloud user experience, and also bring huge economic benefits for cloud computing operators.

云计算利用虚拟机整合技术有效提升了物理资源的利用率，但同时带来了共享资源的竞争和干扰问题，进而导致应用程序的性能异常。如何对云计算系统中存在的性能异常问题进行有效的检测和精准的定位，使其更好地适应云计算系统的大规模、高复杂度、资源共享等特征，提高云计算系统的可靠性，是一个非常有意义的课题。本项目拟围绕当前云计算系统面临的性能异常诊断的实时性差、检测精度低、开销大等问题，从异常特征提取及跨层次相关性感知的异常检测、程序请求路径追踪及基于依赖图动态构建的异常源概率推理等方面开展研究，并构建一个高干扰共享云系统性能异常诊断原型系统。研究成果将能提升云计算系统的可靠性，增强云用户体验，同时也可为云计算运营商带来巨大的经济收益。

本项目围绕高干扰共享云系统对稳定性和健壮性的重大需求，针对共享云计算系统复杂性、动态性、共享性、实时性等特征，开展了高干扰共享云计算系统的异常检测与定位的关键技术研究，在共享云系统中的负载特征分析、性能干扰量化、异常检测机制、异常定位方法等方面取得了重大突破，形成了一套高干扰共享云系统的异常检测机制、异常定位方法及监控预警系统，为保证云服务的正常运行提供了技术支持，并提升了共享云系统的稳定性和健壮性。这些成果将成为促进共享云系统不断深入发展的推动力，将大力促进国家云服务基础设施建设，并满足国家信息安全的重大需求，具有重要的理论和实际意义。.本项目在实施过程中，共发表SCI/EI论文28篇，其中在IEEE Transactions on Cloud Computing等本领域主流国际期刊发表论文4篇，在IEEE CLUSTER、ICPADS等国际会议上发表论文24篇，并获最佳论文奖2篇。申请发明专利共22项，其中PCT国际专利6项，国内发明专利16项，已获得授权5项，获得软件著作权2项。培养研究生14名，其中已毕业硕士4名，在读博士2名、硕士8名。