i2Graph: An Incremental Iterative Computation Model for Large Scale Dynamic Graphs

Zhuo Tang, Mengsi He, Li Yang, Zhongming Fu

Abstract: Due to the rapid changes in graph data sets, the mined information will quickly become obsolete, thus the entire data set needs to be re-computed from the beginning, which will result in the waste of computing time and resources. To reduce the cost of such computations, this paper proposes a model called i2Graph to support incremental iterative computation for dynamic graphs. Different from the way of traditional iteration, i2Graph executes the graph algorithm by reusing the results of the previous graph and performs computation on parts of the graph that has changed. i2Graph contains two components: (1) an incremental iterative computation model to improve the execution efficiency of the iterative graph algorithm; and (2) an incremental update method to accelerate the iterative process within the iterative graph algorithm. It is implemented based on Spark GraphX, a popular parallel and distributed computing framework for large-scale graph processing. Experiment results verify the performance advantages of i2Graph model when performing some iterative graph algorithms on the dynamic graph, compared with the traditional iteration.