A Research And Implementation To Events Connection And Its Label Self-generating Deep Revealing Technology

Any accident of cloud service may lead to disputes between service providers and customers.An effective cloud service accountability mechanism is a necessary means to solve disputes.However,the key premise for the effectiveness of this method is to reveal the abnormal dependence connection between service events,which provides the abnormal connection foundation for the judgment of violation and the source of failure of the accountability mechanism.Current research on cloud service event connection discovery relies heavily on manual pre-labeling,which is labor-intensive and inefficient.Most of its steps are first manual interpretation of data and then manual annotation,another idea is,first machine annotation and then manual interpretation.Therefore,the deep revealing of event connections based on label self-generation has become a research topic.Firstly,this paper points out that event connection still presents unknown discrete characteristics in data,but its discovery method relies on subjective feeling and manual labeling.This causes a difficult problem to be studied urgently in the discovery of event connection.There are disputes between unknown characteristics and current subjective representation.Unknown features are controversial and methodological challenges in manual labeling and machine labeling.Event connection depth found that there are monitoring challenges for manual and machine signatures.Secondly,aiming at the above problems,we study and find that cloud service event connection presents discrete high-dimensional nondeterministic characteristics in data.Based on this,we propose to use 9-dimensional label space to characterize these characteristics,and propose a self-generating strategy based on density clustering and an event-connected machine label filtering strategy.Based on this,we propose to use 9-dimensional label space to characterize these characteristics,and propose a self-generating strategy based on density clustering and an event-connected machine label filtering strategy.Thirdly,based on the label self-generating strategy,we build and construct a label self-generating mathematical model and its method of event connection.Based on the above filtering strategy,we construct a depth discovery method of event connection,and then reconstruct the semi-supervised depth discovery model of label self-generating and event connection.Fourthly,based on the events of "label from generation" and "depth discovery" two new methods,we construct the two key algorithm,and the calculation of the frame model,based on the structure and design of events a label from the frame type depth found algorithm,is given after the above three algorithms complexity of control,It provides theoretical upper limit basis for analyzing the effectiveness of LAGREEN technology.Fifthly,we design and implement the prototype system of label self-generated event connection depth discovery technology,and test the function satisfaction and performance effectiveness of the prototype system according to test cases.There are three new works in this paper.In view of the challenge of the deep implication of multi-party characteristics of events and the visual representation of event connection,we analyze that the current understanding of event connection is one-sided,that is,the anomaly of event and its connection is not only reflected in the event state and its "state pair",but also in the load,duration and retry.The event payload(resource usage)determines whether the event runs properly.At the same time,the abnormal load of the event will also show up in the run time and run retry.Load and duration,and even retry,are deep continuous values that are hard to intuitibly enumerate manually.All of the above factors may represent the anomalies of events and their connections.Therefore,we propose to use nine-dimensional label space to comprehensively characterize the characteristics of event connections.However,some deep characteristics of events that cannot be intuitively recognized(such as "load abnormal" and "long tail" characteristics of events)need to rely on the understanding of machine deep fitting and labeling.Therefore,we propose a self-generating label method based on density clustering to realize machine self-labeling.However,both machine labeling and manual labeling are prone to errors or one-side labeling,so it is necessary to repeatedly screen and correct its misconceptions to approach the real cognition.We present a semi-supervised multi-label connection discovery method,and eliminate and correct the misconceptions through the label screening strategy.The theoretical analysis and experimental results show that the Label self-generating deep revealing for event connection(short for LAGREEN)system proposed by us is capable of label self-generation,uncertain event connection discovery,event network construction,etc.Its functional integrity can reach 0.875,and the system can accurately identify deeper anomaly connection(such as long tail abnormal connection,etc.)with an accuracy of 99.75%?99.97%.