CAJ | 학술논문

通过引入流强度并搜索全部流强度量测之间的不变关系，将大型分布式信息系统建模成一个不变网络，其中节点代表流强度量测，边代表不变关系。当系统发生故障时，会导致不变网络中与故障点相关的边发生中断，且由于故障会在监测数据中进行传播，从而导致多条边发生中断以及多个节点出现异常，加大了系统故障检测的难度。为此，设计了 gRank＋算法，根据节点测量的异常水平对不变网络中的节点进行排序，从而实现系统故障的快速检测。最后使用3组综合数据集通过准确率、召回率和增益值3个指标来验证 gRank＋算法的有效性和优越性。
통과인입류강도병수색전부류강도량측지간적불변관계，장대형분포식신식계통건모성일개불변망락，기중절점대표류강도량측，변대표불변관계。당계통발생고장시，회도치불변망락중여고장점상관적변발생중단，차유우고장회재감측수거중진행전파，종이도치다조변발생중단이급다개절점출현이상，가대료계통고장검측적난도。위차，설계료 gRank＋산법，근거절점측량적이상수평대불변망락중적절점진행배서，종이실현계통고장적쾌속검측。최후사용3조종합수거집통과준학솔、소회솔화증익치3개지표래험증 gRank＋산법적유효성화우월성。
By introducing flow intensity and searching invariant relationships among all flow intensity measurements,large-scale distributed information system is modeled as an invariant network,in which a node denotes a flow intensity measurement and a link indicates an invariant relationship.When a failure happens on a node in the invariant network,it will cause the links related to the node to be broken.More-over,the failure usually propagates among monitoring data,therefore more nodes will be abnormal and more links will be broken,which increases the difficulty of fault detection.This paper proposes an algo-rithm named gRank+ to rank nodes in the invariant network according to the anomaly levels of nodes, which will realize the rapid detection and location of system failure.Finally,the effectiveness and superi-ority of the proposed method is illustrated by using three synthetic data sets in precision rate,recall rate and gain value.