Vectorising k-Truss Decomposition for Simple Multi-Core and SIMD Acceleration

In this paper we tackle truss decomposition of large graphs, which is one of the popular tools for discovering dense hierarchical subgraphs in social and web networks; such subgraphs form the basis of community discovery, one of the cornerstones of modern graph analytics. Our goal is to offer a simple vectorisation approach which can be easily implemented in widely popular Python vector libraries, such as NumPy. This way has two advantages: (1) non-experts with basic knowledge of Python can implement our algorithm, and (2) they can obtain multi-threaded and SIMD parallelism “for free” without them needing to know about computer architecture or sophisticated C++ libraries for multi-threaded processing. We believe this is an important paradigm setting approach that opens the way for applying similar techniques to other problems that might seem at first remote to vectorisation and/or parallelisation.