改进FP-Growth算法下云服务器故障诊断研究

Study on Fault Diagnosis of Cloud Server under Improved FP-Growth Algorithm

大规模云服务器的资源类型较多,数据量巨大,因此云计算环境下服务器参数易发生异常问题,导致服务器产生故障。提出基于改进频繁模式增长(Frequent Pattern Growth, FP-Growth)算法的云服务器故障诊断方法。优先建立支持度函数完成服务器各项参数对应支持度的映射。采用关键字筛选方法将频繁项划分为关键项表和非关键项表两个部分。应用改进的FP-Growth算法,设定参数阈值规则,获取各个服务器故障向量变量的子集,以此为依据组建不同故障变量的条件频繁模式树(Frequent Pattern Tree, FP-tree),通过各个数据子集挖掘出的频繁项集求并集,得到包含全部故障诊断信息的频繁项集,最终完成云服务器故障诊断。经实验测试结果表明,所提方法能够在5ms内完成云服务器多类型故障诊断,且故障诊断精度为90%~100%。实验结果证明,上述方法具有可靠的应用性能。

Large scale ECS has many types of resources and huge amounts of data. Therefore, server parameters in cloud computing environments are prone to abnormal problems, leading to server failures. In this paper, a method of fault diagnosis for cloud server based on improved FP-Growth(Frequent Pattern Growth) algorithm is proposed. At first, the support function was established to complete the mapping of the degree of support corresponding to the parameters of server. Second, frequent items were divided into two parts by keyword filtering method, namely the tables of key items and non-key items. On this basis, the improved FP growth algorithm was applied to setting parameter threshold rules, thus obtaining the subsets of fault vector variables of each server. Based on this, a conditional FP-Tree(Frequent Pattern Tree) of different fault variables was constructed. Moreover, the union was obtained by the frequent itemset mined from each data subset, so that the frequent itemsets containing all fault diagnosis information were obtained. Finally, the fault diagnosis was completed. Experimental results prove that the proposed method can complete the multi-type fault diagnosis of cloud server within 5ms, and the accuracy fault diagnosis is between 90% and 100%. Therefore, this method has reliable application performance.