MDACCER: Modified Distributed Assessment of the Closeness CEntrality Ranking in Complex Networks for Massively Parallel Environments

Abstract

We propose a new method derived from DACCER (Distributed Assessment of the Closeness CEntrality Ranking): the modified DACCER (MDACCER), for assessing traditional closeness centrality ranking. MDACCER presents a relaxation that allows it to take advantage of massively parallel environments like General Purpose Graphics Processing Units (GPGPUs). Traditional DACCER proposal assesses Closeness centrality ranking in a limited neighborhood using only information around each node at low computational cost and capability to be executed in a distributed environment. Despite all the advantages, DACCER presents some difficulties in GPGPUs programming model that increases its computational cost at this particular environment. In contrast to the poor performance of DACCER on GPGPUs, experimental results demonstrate MDACCER is as simple and efficient as DACCER to assess Closeness centrality ranking in complex networks and moreover it does not have the same bottlenecks in GPGPUs computing about memory usage and time complexity. We performed MDACCER for some synthetically generated networks, specifically Barabási-Albert ones and results indicate MADCCER correlates Closeness centrality ranking almost as well as DACCER does with lower computational costs.