Characterising Dependency in Computer Networks using Spectral Coherence

TL;DR

This paper explores applying spectral coherence methods from neuroscience to characterize dependencies in computer network activity, revealing structured dependencies that could help differentiate network behaviors.

Contribution

It introduces the use of spectral coherence to analyze device messaging dependencies in computer networks, adapting techniques from neuroscience.

Findings

Most devices show little dependency between messaging channels.

Detected coherence is often highly structured.

Coherence analysis can potentially discriminate network activity types.

Abstract

The transmission or reception of packets passing between computers can be represented in terms of time-stamped events and the resulting activity understood in terms of point-processes. Interestingly, in the disparate domain of neuroscience, models for describing dependent point-processes are well developed. In particular, spectral methods which decompose second-order dependency across different frequencies allow for a rich characterisation of point-processes. In this paper, we investigate using the spectral coherence statistic to characterise computer network activity, and determine if, and how, device messaging may be dependent. We demonstrate on real data, that for many devices there appears to be very little dependency between device messaging channels. However, when significant coherence is detected it appears highly structured, a result which suggests coherence may prove useful for discriminating between types of activity at the network level.