The first key insight energy-efficient virtualisation is the need for and utility of interfaces between the virtualization layer controlling physical hardware and the operating systems running applications in guest virtual machines.

The purpose of such interfaces is twofold:

1. To inform hypervisors of the power states acceptable to operating systems and thus to the applications they run.
2. To provide implicit or explicit feedback to operating systems about the efficacy of their internal management strategies.

Interestingly, both implicit and explicit feedback can be used, implying that effective power management is possible for guest operating systems whether or not they are aware of running in virtualized environments. However, advantages in terms of power efficiency exist for aware versus unaware systems.

The second insight is that there is a need for new virtualization-level management abstractions. In terms of power management, support is needed for multiple tasks. There must be abstractions for exchanging information about and coordinating management actions across multiple data center machines, subsystems, or partitions (sometimes also called zones) because current hardware is constructed to take per platform and per zone management actions.Without coordination, it is difficult to attain global goals like the enforcement of power caps on zones or entire data centers, and individual decisions can lead to undesirable actions, such as multiple machines being run in lower-power states rather than entirely idling one machine and fully utilizing the other. This is because typically idling machines or turning them off saves more power than running them at lower-power states.

A third insight is that virtualization in fact presents new opportunities for power management, not only in terms of the ability to coordinate power management across multiple machines but also because the virtual power states presented to systems and applications need not be identical to actual power states offered by hardware (which tend to be limited).

Instead, the virtualization layer can use its own innovative mechanisms to create appropriate power states for application use, including those that hide hardware heterogeneity from systems. The last is particularly important in large-scale data centers, where multiple generations of computing hardware coexist in an ever-present rhythm of machine acquisition, deployment, use, and removal.

A question of note is the role and importance of virtualization in high performance computing, in fact, answering this question gives rise to several important new directions in research on effective system power management. Other challenges for power management are due to the increasing fungibility of facilities and applications embedded in virtual machines. Green Cloud Computing has the potential to substantially increase the degree of dynamics experienced in data center systems, so that we must also consider the energy used in dealing with these dynamics (e.g. energy consumed by virtual machines migration) versus just considering steady-state application power consumption.

Unfortunately, in large-scale clouds, operators not be aware of which applications run on which machines at what times, so that it becomes hard or impossible to manage machines to continue to achieve energy-efficient goals. As a result, needed are methods for run-time application recognition and identification without compromising applications' privacy concerns and without requiring applications to undergo onerous modification before they can be run in clouds.