Approaches to Network Management

Managing Computer networks can be a reactive process, set in motion by one or more indicators of an existing problem, or it can be a predictive process, initiated by indicators of the potential for problems in the near future. It is better to predict and avoid network faults (when possible) than it is to detect and repair faults once they occur. This approach is called network steering because the network manager tries to steer the network away from potentially dangerous interactions. Network steering distributes the network manager's work over time, freeing resources for unpredictable faults when they arise.

Consider a trap message generated in response to some feature of a managed object's state exceeding a threshold, such as the number of packets dropped by a router due to a lack of buffer space. It may be that values of that feature as they change over time are correlated with other features of the same object's sate or with features of the state of other objects in the network. If one could find such correlations and use them to predict future states of managed objects, then it would be possible to intervene before the threshold is exceeded and avoid the pathological state that would generate a trap. Note that predicted and existing faults are handled in much the same way. The isolation, diagnosis, and remediation phases following predication or detection of a fault are the same and the same mechanisms can be used in both cases. The advantage afforded by a predictive component is that problems are solved before they reach significant levels, thereby keeping the operation and performance of the network more stable.