SMP clusters are one of the most common HPC platform used by scientific applications. The nodes of SMP cluster contain several computing elements. Scientific applications may be executed over a large number of such nodes introducing complex communication behaviors. Using for instance MPI, communications on a same node with a common interval time create concurrent accesses to resources of nodes. On SMP nodes, concurrent access implies resource sharing depending on the underlying network architecture and the MPI implementation used. This paper presents a model to predict communication times of simultaneous MPI communications over SMP clusters. This model considers the concurrency over resources of nodes and network predicting accurately communication time for many communications in conflict.