Abstract

In large-scale computer networks, the isolation of the primary failure source is a challenging task. This article presents a proactive network fault diagnosis approach based on graph theory. Compared with other approaches, the manager of network management system checks the status of the managed devices actively rather than receive messages from those objects passively. The salient feature of this approach is that the possible failure sources, including the real one, can be computed precisely and quickly without any alarm historical information or strict assumptions. This approach does not introduce much processing complexity by taking full use of matrix and Boolean operations. To test and evaluate our proposed algorithm, it is implemented in Java and tested in a real large network environment. The experiment results show that this approach is not only efficient but also scalable on fault identification in large-scale computer networks.

Keywords: Connectivity; Connectivity structure; Graph theory; Network management; Fault identification; Simple network management protocol; Internet control messages protocol

Article Outline

1. Introduction

2. Network modeling

3. The fault identification approach

4. Fault cases studying

4.1. A vertex is inaccessible whose degree is larger than 1

4.2. A vertex of degree 1 is unreachable

4.3. Multiple vertices in a connected component are inaccessible

4.4. Multiple vertices are inaccessible which are in multiple connected components

5. Automated fault effect analysis algorithm

5.1. Possible fault elements combination and its effect

5.2. Judging whether fault effects are the consistent

5.3. Algorithm for computing fault elements combination

6. Emulations in a large-scale network

7. Conclusions

Acknowledgements

References

Vitae