Network root fault location based on network topology and alarm

Large Internet service platforms involving hundreds of inter-system calls generate a large amount of alarm data every day. How to use the network topology information and alarm data to analyze the alarm in a timely and effective manner, and finally give the effective alarm and the suspected root cause, is the main challenge facing the network operation and maintenance. This paper studies a kind of solution. First, we preprocess the output sequence of a long alarm system cluster. And then judge whether there is a root fault by Support Vector Machine. At the next stage, employee a well-prepared Bayesian network to compute the highest probability of fault types, combined with filtering rules to get the final conclusion. The method is lightweight and efficient, which has been verified by experiments.