Network-Aware and Unsupervised Diagnostics for Latent Issues in Network Management Databases
Li Chen, Hua Shao · 2019
Network management database (NMD) is essential in modern large-scale networks. Operators rely on NMD to provide accurate and up-to-date data, however, NMD---like any other databases---can suffers from latent issues such as inconsistent, incorrect, and missing data. In this work, we first reveal latent data issues in NMDs using real traces from a large cloud provider, Tencent. Then we design and implement an diagnostic system, NAuditor, for unsupervised identification of latent issues in NMDs. In the process, we design a compact and graph-based data structure to efficiently encode the complete NMD as a Knowledge Graph, and model the diagnostic problems as unsupervised Knowledge Graph Refinement problems. We show that the new encoding achieves superior performance than alternatives, and can facilitate adoption of state-of-the-art KGR algorithms. We also have used NAuditor in a production NMD, and found 61 real latent issues, which all have been confirmed by operators.