Cognitive Networks: Towards Self-Aware Networks

5.3. Tasks in Cognitive Networking

The vision for the Knowledge Plane ^{[] [] []} describes a number of novel capabilities for computer networks. This section reviews three capabilities that the vision assumes in terms of the cognitive functionalities that are required. These include anomaly detection and fault diagnosis, responding to intruders and worms, and rapid configuration of networks.

5.3.1. Anomaly Detection and Fault Diagnosis

Current computer networks require human managers to oversee their behavior and ensure that they deliver the services desired. To this end, the network managers must detect unusual or undesirable behaviors, isolate their sources, diagnose the fault, and repair the problem. These tasks are made more challenging because large-scale networks are managed in a distributed manner, with individuals having access to information about, and control over, only portions of the system. Nevertheless, it will be useful to examine the activities in which a single network manager engages.

The first activity, anomaly detection, involves the realization that something unusual or undesirable is transpiring within the network. One possible approach to this problem, which applies recent advances in Bayesian networks, is to formulate it as a density estimation problem. Individual components, larger regions of the network, or, at some level, the entire Internet could be modeled as the joint probability distribution of various quantities (queue lengths, traffic types, round-trip ...