云系统软件老化的双层概率模型及关键技术研究

Software aging is caused by triggering "aging related bug" as well as error propagation, causing system resource exhaustion, error accumulation in the long-running system, its thus leads to increasing failure rate and performance degradation. Triggering "aging related bug" is non-deterministic, and it has close relationship with the environmental factors such as load, resource exhaustion and system running time. Cloud system needs high dependability, which is the current hot research topic for the cloud system. However, the phenomena of software aging, which make cloud system performance degradation, unavailability, and even crash, is not valued highly by the industrial system. It will cause great loss if it happens in the cloud system.. In this project, we will start with the realistic happened cases for cloud system, and based on the fundamentals of software aging as well as the characteristics of cloud system, and focus on the key technologies in two granularity software aging probability models for the cloud system, to optimize the software rejuvenation time, improve the system availability, guarantee the response time and reduce the loss packet probability. In this project, we will investigate the key technologies as follows. Avoid and tolerate the state space explosion, the thresholds for two-layer VM as well as VMM, the detection algorithm for response time aging. . In this project, we employ the cross-validate approach among probability model, experiment and simulation, to solve cloud system aging problem comprehensively. We will also make use of two-layer probability model, threshold determination methods and response time aging detection technique to build cloud- based software monitoring tools, to ease the cloud system monitoring and testing tasks.

软件老化是由于“老化”缺陷的触发、错误的传播，长时间运行导致系统资源的消耗、错误的累积，造成软件失效率增加、系统性能衰退。老化缺陷的触发具有不确定性、与负载、系统运行时间等密切相关。云系统的可信性是当前研究的热点，但软件老化现象作为云系统性能衰退、可用性下降的重要因素，未引起业界足够重视，一旦发生会造成重大的经济损失。. 本项目从实际的案例出发，根据软件老化的机理和云系统的特点，以及监测器的行为对系统的影响，研究云系统双层(虚拟机VM、虚拟机管理器VMM)的老化，并兼顾监测器行为的概率模型及关键技术-避免和容忍模型状态空间爆炸问题、双层老化阈值确定方法、响应时间老化检测算法。. 本研究采取概率模型、实验、仿真交叉验证方法，综合地解决受老化影响云系统可信性的问题。本项目将利用双层概率模型理论、老化阈值确定方法和响应时间老化检测算法，针对云系统的特点和软件老化实现云服务监测工具。

云系统的可信性是当前研究热点，软件老化现象作为云系统性能衰退的重要因素，未引起业界足够重视，一旦发生会造成重大经济损失。在本项目的支撑下，课题组针对影响云平台安全的因素展开研究并取得了多项技术突破。包括基于两粒度检验的软件再生技术，SQLi测试技术及软件中原子性违规问题的修复技术。.基于两粒度检验的软件再生技术是一种闭环控制技术，用于减轻两粒度（应用程序和操作系统）软件老化的负面影响。我们基于系统条件构造马尔可夫再生过程模型，通过故障注入实验获得应用软件和操作系统的性能下降率。实验验证两粒度软件的再生比传统的单粒度软件再生更有效。在本实验中，与单粒度再生相比，当使用两粒度软件再生时，系统的不可用性和整体丢失率分别降低17.9％和2.65％。.针对Web系统常见的SQLi漏洞，我们提出组合突变方法生成SQLi的测试集技术。该测试方法适用于t-way和可变强度组合测试。实验结果表明，该方法可有效发现具有多个过滤规则的SQLi漏洞。同时，我们从测试用例优先级的角度出发，提出一种更有效的SQLi发现技术ART4SQLi，在三个开源易受攻击的基准上进行实验，结果表明ART4SQLi可以在成功完成注入之前减少SQLi尝试次数，比常规随机测试方法提高了26％以上的效率。.针对难以解决的原子性违规问题，我们提出xFixer自适应解决原子性违规问题。xFixer通过消除新的锁定顺序避免引入死锁。并且xFixer可以对带有锁的原子性违规实施两种修复策略。使用xFixer对15个使用原子性违规的操作进行实验，结果表明xFixer正确地修复了所有15个原子性违规，而没有引入死锁。.本项目共在ToR、Vehicular communications等期刊以及FSE、SATE等国内外高水平会议上发表8篇论文，并申请9项发明专利（1项被授权）。基于上述研究成果，课题组研发了分布式实时云监控系统，它采集云平台运行时的关键性能指标，实时检测并预测异常。该软件在某军用项目中使用。本项目研究成果有助于实施云系统性能测试和运行监控，对提高云系统的可用性、降低系统的维护费用并保障云用户的响应时间性能提供理论依据和方法。