Ontology mapping systems are used in different areas of applications, including the real time ones. In order to increase their performance various approaches have been proposed and a large number of mapping systems were developed, but their performance fails when it comes to large ontologies. In the ontotogy mapping process the similarity matrix calculus takes about 50% of the entire execution time. We focused on this calculus and we proposed a parallel algorithm for computing the linguistic similarity matrix of two ontologies. This algorithm has been tested in different configurations: one or more computing nodes, different number of running processes. Different specializations of running processes were taken into consideration: either all for computing, or one for distributing and managing the results of the tasks and the rest for computing. For implementation we took into account the Java MPI, the MPJExpress project was used as middleware for messaging communication in Java, and we evaluated the time of message passing. The resulted module for similarity matrix computing was inserted in Falcon-AO, one of the systems with the best execution time for medium size ontology, in order to improve its overall mapping performance.