Buffered Streaming Graph Partitioning

Partitioning graphs into blocks of roughly equal size is a widely used tool when processing large graphs. Currently, there gap observed in the space available partitioning algorithms. On one hand, are streaming algorithms that have been adopted to partition massive graph data on small machines. In model, vertices arrive at time including their neighborhood, and then be assigned directly block. These can huge quickly with little memory, but they produce partitions low solution quality. other offline (shared-memory) multilevel high-quality also need machine enough memory networks. this work, we make first step close by presenting an algorithm computes significantly improved using single setting. First, adopt buffered model which more reasonable approach practice. element store buffer nodes alongside edges before making assignment decisions. When our receives batch nodes, build represents already present structure. This enables us apply turn, cheap machines, compute much higher quality solutions than previously possible. To graph, develop optimizes objective function has shown effective for Surprisingly, removes dependency number k from running compared previous state-of-the-art. Overall, computes, average, 75.9% better Fennel [ 35 ] very size. addition, values becomes faster .