A parallelization based on MPI and OpenMP of an algorithm that evaluates and counts all the possible communities of a graph is presented. Performance results of the parallelization of the algorithm obtained on a cluster of workstations are reported. Load balancing was used to improve the speedups obtained on the cluster. Two different kinds of load balancing approaches were used: One that involved only MPI and a second one in which MPI and OpenMP were combined. The reason for the load imbalance is described.