Analysis of partitioning strategies for graph processing in bulk synchronous parallel models.

Vertex centric computation implemented with a Bulk Synchronous Parallel (BSP) model is becoming a popular choice to analyze huge graphs. In this paper, we study the impact of the graph partitioning strategies for BSP by simulating different partitions of the vertices of the graph. We also consider the impact of replication and heuristic strategies to balance the management of the replicas. We observe that the edge cut of the resulting graph is not a good indicator of the performance of the global system. Moreover, we find that the choice of an efficient partition strategy is very dependent on the locality of the graph operation.