Influence maximization is the problem of how to pick the best set of seed vertices from a social network so that the picked vertices offer largest influence in the network. Traditional Monte Carlo solution is not feasible for large scale social networks due to its complexity. Therefore in this work, we attempt to design a fast algorithm based on Map-Reduce, i.e., a distributed computation framework. The basic idea is to partition the network and identify the local maximum influence vertices in each partition simultaneously. However, to find the local maximum vertices, it is necessary to exchange the reachable information with their neighbor partitions. In order to save the communication cost, we carefully aggregate the information and process them by batch. Besides minimizing the communication cost, our experiments demonstrate that our approach is superior to existing approaches in terms of Clock Time and Machine Time