一种面向可靠云计算的自适应故障检测方法

提出了一种保障云系统可靠性的自适应故障检测方法。首先基于步进指标搜索算法和指标空间分离技术进行云指标提取, 得到最能刻画云行为和健康状况的最大关联性标准和最小冗余度标准, 然后使用最小封闭球体将云指标数据点从数据空间映射到内核空间进行降维, 最后将低维数据输入故障检测器。故障检测器在内核空间产生一个能够容纳数据的最小封闭球面, 确定潜在故障, 如果以前故障历史无法获得, 则故障检测器通过寻找与其他状态存在明显不同的云健康状态来确定故障。当检测结果被云运营商验证后, 要么被判定为真实故障, 要么被判定为正常状态（虚警）。实现了一种故障检测原型系统, 并在校园云计算环境下展开实验。实验结果表明, 与其他现有故障检测技术相比, 该方法的检测更高效、更准确。

Adaptive fault detection method for reliable cloud computing

This paper proposed an adaptive fault detection method to protect the reliability of cloud system. Firstly, it exacted the cloud metric using the incremental metric search algorithm and the metric space separation technology, which could obtain the best metric for characterizing the cloud behavior and the health status such as the maximal relevance criterion and the minimal redundancy criterion. And then mapped the data point of cloud metric into the kernel space to reduce the dimensionality by using the minimum enclosing sphere formulation. Finally, input the low-dimensional data into the fault detector, which generated a smallest enclose spherical with accommodating data in the kernel space to identify the potential failure. If the fault history could not be obtained, the fault detector would find the other states different from the cloud health status to identify the failure. The cloud operators verified and confirmed the detection results as either true failures or normal states (false alarms). It have implemented a prototype of the fault detection system and conducted experiments in an on-campus cloud computing environment. The experimental results show that the method can achieve more efficient and accurate failure detection than other existing schemes.