Sufficiently adequate dynamic models of the functioning of complex systems are characterized by high computational complexity, which leads to a significant complexity of optimization procedures. Therefore, the solution of the problem of combinatorial optimization by a complete search of possible solutions in practice is unacceptable. The advantages of evolutionary search as a method of combinatorial optimization of the structure of the information and computing system is the possibility of various ways of setting the target function and types of optimization variables, as well as in the use of probabilistic, rather than deterministic rules for finding solutions. A method of formalizing the structure of a heterogeneous information and computing system is proposed, which takes into account its hierarchical-network structure. The presented approach allows by selecting rational parameters of the genetic algorithm and using the stochastic fitness function with a variable coefficient of variation to achieve a satisfactory speed of its convergence with a large dimension of the task. Examples of application of this method to the search for a rational structure of the computing system are given.