Synthetic diagnostics for global computer networks and fusion power experiments

The fast, efficient and effective automated monitoring of complex, interconnected computer network systems for performance optimisation and security is an increasingly necessary task for network administration. With the increasing complexity and connectivity of computer network topologies, even monitoring has become a difficult task. A significant part of the problem is the lack of a quantitative understanding that describes the relationship between network parameters, and the expected response of the network to coarse measures of external stimuli (e.g. threat levels). In short, the network “physics” is unknown.

This project aims to characterise the properties of network-packet captures (both batched and streaming), and use a combination of statistical techniques (such as low-order moments), Fourier and higher order spectral methods and correlation analysis techniques to develop candidate reduced “forward-models” for the network parameters. Outcomes would inform a longer term project that hopes to develop tomography for network properties in response to network shifts (e.g. configuration changes, malicious attack, malfunction).

The project will also develop a skills set in the generation of synthetic diagnostics, which are portable to other fields of research, such as integrated modelling of data and models in fusion power experiments. Such experiments have a large collection of heterogeneous diagnostics, and the integration of these diagnostics with physics models is an outstanding integrated modelling challenge.