The parallel computing of node centrality based on GPU

Abstract

Many systems in real world can be represented as network, and network analysis can help us understand these systems. Node centrality is an important problem and has attracted a lot of attention in the field of network analysis. As the rapid development of information technology, the scale of network data is rapidly increasing. However, node centrality computation in large-scale networks is time consuming. Parallel computing is an alternative to speed up the computation of node centrality. GPU, which has been a core component of modern computer, can make a large number of core tasks work in parallel and has the ability of big data processing, and has been widely used to accelerate computing. Therefore, according to the parallel characteristic of GPU, we design the parallel algorithms to compute three widely used node centralities, i.e., closeness centrality, betweenness centrality and PageRank centrality. Firstly, we classify the three node centralities into two groups according to their definitions; secondly, we design the parallel algorithms by mapping the centrality computation of different nodes into different blocks or threads in GPU; thirdly, we analyze the correlations between different centralities in several networks, benefited from the designed parallel algorithms. Experimental results show that the parallel algorithms designed in this paper can speed up the computation of node centrality in large-scale networks, and the closeness centrality and the betweenness centrality are weakly correlated, although both of them are based on the shortest path.

Keywords

• node centrality
• parallel computing
• cuda