According to one embodiment, availability information describing virtual machines running on physical machines is accessed. The availability information associates each virtual machine with a physical resource used by the virtual machine. Use by the virtual machines is determined from the availability information. Availability of the physical resources is determined according to the use. Server load is balanced according to the availability of the physical resources. According to another embodiment, the following is performed until a load is accommodated: selecting a server having a load that is less than an expansion threshold; loading the server until the expansion threshold is reached; selecting a next server having a load that is less than a next expansion threshold; and loading the next server until the next expansion threshold is reached. Load of a low load server is determined to be below a contraction threshold, and the low load server is drained.