The optimal mapping of tasks of a parallel program onto nodes of a parallel computing system has a remarkable impact on application performance. We propose a new criterion to solve the mapping problem in 2D and 3D meshes that uses the communication matrix of the application and a cost matrix that depends on the system topology.We test via simulation the performance of optimization-based mappings, and compare it with consecutive and random trivial mappings using the NAS Parallel Benchmarks. We also compare application runtimes on both topologies. The final objective is to determine the best partitioning schema for large-scale systems, assigning to each application a partition with the best possible shape.