G-FDDS: A graph-based fault diagnosis in distributed systems

With the rapid development of Internet technologies such as cloud computing and big data, the scales of distributed information systems in big companies have grown to enormous sizes. Automatic detection and diagnosis of system faults in the large-scale information systems is complicated and important in both practice and research. In this paper, we propose a Graph-based Fault Diagnosis approach in Distributed System (G-FDDS), in which availability testing records of distributed systems are represented by a multi-relational graph in order to find out latent root causes of the failures of availability testing. Vertices represent non-repeated failed availability testing records and edges indicate multiple relations between these records, then a clustering method is proposed to group similar vertices into fault causes. Our approach is more appropriate for mining accurate fault causes both in the simulation and real datasets comparing with other commonly used methods such as statistical methods.