Research On Control Log Graph Mining For System Log

With the rapid development of cloud computing,modern software developers will choose to build their own systems,and the system logs are printed in different forms due to the different writing styles of programmers,and it is difficult for non-professionals to understand the meaning.Although the log has the above characteristics,the log is the only source of information for the system to record its own operating status,and it contains the internal execution status of the system.Understanding the log is helpful for the operator to fully understand the system's design architecture.Extracting key information from unstructured plain text system logs,and then constructing a control flow diagram representing the system's operation,and further discovering abnormal events in system deployment are important issues that need to be resolved in the computer field.This is called the control flow graph mining process.This paper studies the process of transaction flow graph mining.The main research contents are as follows:(1)This paper first analyzes the time sequence relationship between events in the log sequence and dig out the time lag interval between events,determine the dependency of the event through the condition/unconditional distribution of the event,and determine the overall control flow according to the substructure of the control flow graph Graph generation,through comparison with other control flow analysis methods,verify the effectiveness and feasibility of the algorithm in theory and examples.(2)For the analysis of the variable part of the system log,compare the identifier information that is generally present in the log text,and further analyze the process of the identifier passing information in the different component logs to determine the strong dependency between the events;time series analysis of the log sequence can be obtained Weak dependency.Combining the two pairs of strong and weak dependency events to construct a normally executed control flow graph and comparing it with other control flow graph parsing methods,it proves the effectiveness and feasibility of this method experimentally.(3)Finally,this paper applies control flow graph analysis to an important area of log analysis-anomaly detection.On the one hand,it introduces two common anomaly detection architectures;on the other hand,analyzes the anomaly events obtained by the algorithm in this paper on the enterprise data set,which proves the rationality and effectiveness of the algorithm.