iPregel: Vertex-centric programmability vs memory efficiency and performance, why choose?

The vertex-centric programming model, designed to improve the programmability in graph processing application writing, has attracted great attention over the years. However, shared memory frameworks that implement the vertex-centric interface all expose a common tradeoff: programmability against memory efficiency and performance. Our approach, iPregel, preserves vertex-centric programmability, while implementing optimisations for performance, and designing these so they are transparent to a user's application code, hence not impacting programmability. In this paper, we evaluate iPregel against FemtoGraph, whose characteristics are identical, an asynchronous counterpart GraphChi and the vertex-subset-centric framework Ligra. Our experiments include three of the most popular vertex-centric benchmark applications over 4 real-world publicly accessible graphs, which cover orders of magnitude between a million to a billion edges, measuring execution time and peak memory usage. Finally, we evaluate the programmability of each framework by comparing against Google's original Pregel framework. Experiments demonstrate that iPregel, like FemtoGraph, does not sacrifice vertex-centric programmability for additional performance and memory efficiency optimisations, which contrasts against GraphChi and Ligra. Sacrificing vertex-centric programmability allowed the latter to benefit from substantial performance and memory efficiency gains. We demonstrate that iPregel is up to 2300 times faster than FemtoGraph, as well as generating a memory footprint up to 100 times smaller. Ligra and GraphChi are up to 17000 and 700 times faster than FemtoGraph but, when comparing against iPregel, this maximum speed-up drops to 10. Furthermore, with PageRank, iPregel is the fastest overall. For memory efficiency, iPregel provides the same memory efficiency as Ligra and 3 to 6 times lighter than GraphChi on average.