SLIDE 15 STREAMER: ¡a ¡Distributed ¡Framework ¡for ¡Incremental ¡Closeness ¡Centrality ¡Computa@on ¡
IEEE Cluster’13
Performance ¡Results ¡
2,000 4,000 6,000 8,000 10,000 12,000 10 20 30 40 50 60 Updates per second Working nodes 8 threads, 1 graph/NUMA 8 threads, 1 graph 8 threads, 1 graph/thread 4 threads, 1 graph 1 thread 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 45,000 50,000 10 20 30 40 50 60 Updates per second Working nodes 8 threads, 1 graph/NUMA 8 threads, 1 graph 8 threads, 1 graph/thread 4 threads, 1 graph 1 thread 2,000 4,000 6,000 8,000 10,000 12,000 10 20 30 40 50 60 Updates per second Working nodes 8 threads, 1 graph/NUMA 8 threads, 1 graph 8 threads, 1 graph/thread 4 threads, 1 graph 1 thread 100 200 300 400 500 600 700 800 10 20 30 40 50 60 Updates per second Working nodes 8 threads, 1 graph/NUMA 8 threads, 1 graph 8 threads, 1 graph/thread
456x speedup 497x speedup 316x speedup linear scaling sublinear scaling (will be explained)
(a) amazon0601 (c) web-Google (d) soc-pokec (b) web-NotreDame
15