MIB variable based fault classification: the next step towards proactive management

Faults can be classified using information contained in the MIB variables. It is observed that distinct changes in the MIB data characteristics precede different fault types. Previous work has shown that network faults can be predicted using MIB data. Now with the possibility of classifying faults, network alarms can be associated with specific fault types. Associating a network alarm with a specific network fault is an essential pre-requisite for automated recovery. In our current work we describe the classification of four different faults: network access problems, protocol implementation error, runaway process and file server failures. We also show that a simple discriminant function scheme that accounts for spatial correlations in the MIB data performs better than common majority-voting schemes.