GreenBFS: Space-Efficient BFS Engine for Power-aware Graph Processing

BFS (Breadth-First Search) is a promising killer engine for graph processing and currently has an extremely important role in real-time processing scenarios in everyday life. Unfortunately, the current BFS engine often delivers poor efficiency owing to complex graph representation, and high energy/memory cost. This paper presents GreenBFS, a space-efficient BFS engine for power-aware graph processing. GreenBFS (i) adopts Folding@CSR, a novel compressed sparse row(CSR) format based on folding edges for sorted CSR. Folding@CSR can store graph data using 50% smaller memory compared to the state-of-the-art graph compression; and (ii) proposes a mixed-conservation mode based on hardware and software integration that can attain power conservation by 35.7%. We use both benchmark and real-world graphs to demonstrate the effectiveness of GreenBFS. Firstly, GreenGraph500, a widely accepted benchmark to rank graph processing capability with energy, is used to validate GreenBFS. GreenBFS-based Tianhe Exa-node has won both No.1 in the latest GreenGraph500 ranking (June, 2022) in big data and small data categories. We finally apply GreenBFS to construct graphs, which is much more efficient on memory footprint than that of the state-of-the-art graph representation with lower electric supply.