This article presents a proposal of solution for the problem of optimization of the virtual machine (VM) backup or replication process in architecture with multiple locations where efficient bandwidth usage and maximal tardiness for single VM are objectives. The two step algorithm is considered, where in the first step a set of tasks is partitioned into smaller subsets for load balancing. In the second step - tabu search algorithm is used to minimize weighted sum of tardiness. The paper contains the results of computational experiments on the scalability of the presented method.