This thesis initially concerned the parallelization of the OSSAU software code from CEA/CESTA. The application domain of this software is the vectorized structural mechanics; the code itself is non linear in time and in two or three dimensions. This investigation leads to the conception and the implementation of a parallel high performance software processing chain for the assembly and the resolution of sparse linear systems by direct methods. The final objective is a validation of the OSSAU code for three dimension problems with several millions of unknowns.