Subsystem-level power management in a multi-node virtual machine environment

A computer-implemented method includes capping the amount of power available to each of a plurality of compute nodes, and managing power allocation among subsystems within each of the compute nodes according to the requirements of workloads assigned to each of the compute nodes. The method further comprises reporting an actual performance level and performance capability for each subsystem within each of the plurality of compute nodes, and monitoring parametric data for a particular workload. A target compute node is identified from among the compute nodes, wherein the target compute node would be capable of performing the particular workload if power was reallocated from a first subsystem to a second subsystem within the target compute node. The particular workload is then assigned to the target compute node. Optionally, assigning the particular workload may include migrating the workload to the target compute node from another of the compute nodes.