The increasing emphasis on energy efficiency has resulted in the deployment of power management policies that are tuned for specific workloads or classes of applications.

The parameters that govern the behavior of these policies can be fine-tuned for individual applications to improve power savings while meeting performance requirements. The existence and deployment of these policies help meet application-specific Quality of Service constraints when performing active power management. Therefore, it is highly desirable to continue leveraging these policies under virtualized execution.

Efforts toward performing power management using virtual machine based policies must simultaneously meet constraints of isolation and independence.

Isolation refers to the fact that power management actions performed on behalf of one virtual machine should not adversely affect another, co-hosted, application. Similarly, any attempt to make use of virtual machine specific policies must maintain the independent manageability expected by guests.

To meet these design requirements, an approach is needed where virtual machine management policies convey power/performance trade-offs to the management partition. The information is then used as feedback to authoritative policies in the management partition, which eventually modify resource allocations and power states of physical resources.

The power consumption of a processor is proportional to the product of frequency and voltage squared. Therefore, voltage scaling is required to obtain significant power savings.

However, often there may be fewer voltage rails than available processors, and the effective voltage is limited by the fastest running processor. When the shadow Virtual Power Management states assigned to the virtual processor of distinct virtual machines are different, the ability of Virtual Power Managment rules to utilize hardware states for power savings may be limited.

Soft scaling can provide power benefits because it duty cycles the underlying hardware, allowing it to enter lower-power idle states. Indeed, with the increased emphasis on idle power management, soft scaling can help hardware attain energy-saving states that improve power consumption beyond the hardware capabilities available.