This paper considers cluster systems of data centers containing a definite set of application servers, file servers, data storage systems, and an input-output system that are interconnected by a switching system and communication channels. The goal of research is to increase the efficiency of virtualized cluster systems using a new rational method to distribute virtual machines among the physical elements of data centers. This method is based on an iterative greedy algorithm and a limited enumeration procedure, which improve the performance of a virtualized system due to rational data management. A simulation model for the operation of a virtualized system is developed, and its experimental study is carried out. The approaches presented below determine the required number of cluster nodes in a virtualized system under an unpredictable load rate and an unpredictable level of resources requested by virtual machines.