FastBFS: Fast Breadth-First Graph Search on a Single Server

Abstract

Big graph computing can be performed over asingle node, using recent systems such as GraphChi and X-Stream. Breadth-first graph search (a.k.a., BFS) has a patternof marking each vertex only once as visited and then notusing them in further computations. Existing single-servergraph computing systems fail to take advantage of such accesspattern of BFS for performance optimization, hence sufferingfrom a lot of extra I/O latencies due to accessing no longeruseful data elements of a big graph as well as wasting plentyof computing resources for processing them. In this paper, we propose FastBFS, a new approach thataccelerates breadth-first graph search on a single server byleverage of the preceding access pattern during the BFSiterations over a big graph. First, FastBFS uses an edge-centric graph processing model to obtain the high bandwidthof sequential disk accesses without expensive data preprocessing. Second, with a new asynchronous trimming mechanism, FastBFS can efficiently reduce the size of a big graph byeliminating useless edges in parallel with the computation. Third, FastBFS schedules I/O streams efficiently and can attaingreater parallelism if an additional disk is available. We implement FastBFS by modifying the X-Stream systemdeveloped by EPFL. Our experimental results show thatFastBFS can outperform X-stream and GraphChi in thecomputing speed by up to 2.1 and 3.9 times faster respectively. With an additional disk, FastBFS can even outperform themby up to 3.6 and 6.5 times faster respectively.