In this paper, we introduce a generalization of the vector bin packing problem, where the bins have variable sizes. This generalization can be used to model virtual machine placement problems. In particular, we study the machine reassignment problem. We propose several greedy heuristics for the variable size vector bin packing problem and show that they are flexible and can be adapted to handle additional constraints. We highlight some structural properties of the machine reassignment problem and use them to adapt our heuristics. We present numerical results on both randomly generated instances and Google realistic instances for the machine reassignment problem.