Community detection is essential to various graph analysis applications. Infomap is a graph clustering algorithm capable of achieving high-quality communities. However, it remains a very challenging problem to effectively apply Infomap on large graphs. By analyzing communication and workload patterns of Infomap and leveraging a distributed delegate partitioning and distribution method, we develop a new heuristic strategy to carefully coordinate the community constitution from the vertices of a graph in a distributed environment, and achieve the convergence of the distributed clustering algorithm. We have implemented our optimized algorithm using MPI (Message Passing Interface), which can be easily employed or extended to massively distributed computing systems. We analyze the correctness of our algorithm, and conduct an intensive experimental study to investigate the communication and computation cost of our distributed algorithm, which has not shown in previous work. The results demonstrate the scalability and the correctness of our distributed Infomap algorithm with large-scale real-world datasets.