Speeding up the simulation of propagation processes in large networks using GPU

OData support
Supervisor:
Dr. Gulyás András
Department of Telecommunications and Media Informatics

Dynamic Network Analysis (DNA) is a constantly emerging scientific field which covers exciting areas inside network theory like examining changes, dynamics of social network connections, multi-agent systems of artificial intelligence or molecule- and protein structure networks.

These networks (or from an other point of view, graphs) differ very much from each other both in size and in structure. While a protein structure network consists only of 100 nodes and 1000 links, an extensive social or computer network could contain several millions of nodes and tens of millions of links among them. Analysing their dynamics is a time-consuming and complex task, for which some network analyser programmes were made. Among them in this paper the Hungarian-developed Turbine will be presented. Unfortunately, the performance of these programmes are not satisfactory for analysing big and complex networks; the long execution time makes it difficult to work smoothly.

Today's video cards let their computing power not be used solely for displaying images, but - observing a few special rules - can be capable of running programmes made by us thus we can utilise their full potential and vastly accelerate computations. This method is called general-purpose computing on graphics processing units (GPGPU). From its specific implementations NVIDIA CUDA was chosen.

Huge neural- or social networks are consisted of several millions or tens of millions nodes and links. Even if a programme is able to handle this big graphs, the analysis time can be measured in hours. In this paper it will be presented how can this amount of time be reduced to minutes using video cards.

Downloads

Please sign in to download the files of this thesis.