Numerical models develop with development of computational technique, and this development can include either creation of new models, or modification, improvement of existing ones. Modifications can concern both mathematical part, for example, change of numerical schemes, parameterization of various physical processes, and technical part, for example, adaptation of algorithms for use on other computer systems or parallelization of algorithms. The paper considers the issue of optimization of parallel computations for the ice and ocean model SibCIOM. Optimization consists in constructing a mesh partitioning for which the nodes are evenly distributed across the computational cores and the time spent on exchanges is minimal. The METIS package is used to implement such a mesh partitioning. A conceptual description of the implementation of exchange processes for such partitioning is presented.