Using Community Detection Algorithms to Identify Clusters of Ranks in an MPI Application Based on the Communication Matrix

In the MPI parallel programming model, communication remains the bottleneck that prevents applications to achieve greater performance and scalability. Due to this problem it is important to know the behavior of this communication in each application. For this reason, we propose the use of community detection algorithms to identify from the communication matrix the clusters of ranks that maximize intracluster communication and minimize intercluster communication. The aim of this project is providing another tool to identify how you can improve the performance of a MPI application