Request Tracing And Anomalies Detecting System In Cloud

As a maturing technology, the scale of cloud computing is expanding exponentially.Cloud platform itself has large-scale components and the dependencies of components arecomplex. A large number of external requests flow between the various components ofcloud platform, which has brought a lot of pressure. For request-driven cloud services, it isnecessary to understand the performance state while processing every request in cloud.And it is also necessary to detect and locate the abnormal component, if failure orperformance degradation occurred during request processing. But, the traditional meanscannot resolve these problems.Request tracing and anomaly detecting system is designed to resolve the problemmentioned above. A unique global identifier will be assigned to each external request, andattached to the invoking context during the whole lifecycle of request processing. And atrace entry will be generated with the invoking information including response time,processing state and call dependency of each component. Then according to the identifierrecorded in trace entries, trace entries are delivered into different queue which belong to acertain request path. After sorting all trace entries in each queue, a request tracing path isreconstructed and classified according to call appearance in each path. By analyzing thestructure characteristics of each request path and the execution state in correspondingcomponent, filter out abnormal request paths in which functional exception occurs. Thenuse a clustering algorithm to cluster each component’s response time and locate abnormalcomponent in which functional exception occurs.Functional tests show that the request tracing and anomaly detecting system canobtain each external request’s execution status and response time in various components.And it also can reconstruct the topology graph of each request flow path by analyzingtrace entry of components. Besides，this system can detect not only functional anomaliesbut also performance anomalies. Performance tests show that obtaining and tracingrequest from inside of cloud platform does not cause significant overhead. And theoverhead of response time is maintained at about5%.