With The development of computer systems and networks, the security and stability of them are becoming more and more important. As a way to improve system's dependability, the research on system level ,fault diagnosis has great significance.In this paper, a survey of system level fault diagnosis is briefly introduced. Then a Boolean equation is proposed to represent the problem of fault diagnosis. Four diagnosis models are described as Boolean equations, and some related problems are discussed based on Boolean equations such as optimal diagnosis, all consistent fault patterns, absolute fault-free and faulty machines etc. Next the four models are described as linear equations and then MATLAB can be used in equation diagnosis.Based on greedy algorithm and grouping theory, four probabilistic diagnosis algorithms are presented, and each algorithm has different greedy criterion. Experimental results show that every algorithm has high correctness and low time-complexity of diagnosis. And these algorithms can change into deterministic diagnosis algorithms if t-diagnosable conditions are satisfied. We can also see that these algorithms have better performance than the Majority algorittim and Compete algorithm, which are classic probabilistic algorithm in system level fault diagnosis.A distributed hierarchical diagnosis algorithm is discussed for virtual private networks. This algorithm uses distributed diagnosis on every subnet and transfers information using IPSec protocol. Next a practical broadcast algorithm is implemented in Ethernet. This algorithm primarily uses the broadcast characteristics and can change the centralcomputer dynamically if the original central computer is down. |