Chapter One
It's a story that is all-too-familiar across organizations today. Rising electricity costs, shrinking power supplies, and mounting social and economic pressure to be "green" are forcing organizations across the globe to rethink their IT strategies. For an increasing number of organizations, server virtualization has become a significant means to a power-reducing end. In fact, according to recent research from the Enterprise Strategy Group (ESG), organizations resoundingly think server virtualization will have the single, greatest impact on reducing power consumption in the data center - twice the impact above the development of more energy-efficient physical server technologies.

Why? Simple: Server virtualization enables the consolidation of physical server environments, which means fewer physical servers are needed to run the same number of applications. While the ratio of virtual machines to physical servers does vary depending on the server virtualization technology used to create the virtual environment, the type and scalability of deployed physical servers and user-defined service-level agreements (SLAs), etc., 5:1 reduction is typical. The bottom-line savings resulting from powering and cooling as many as 80% fewer physical servers is significant for most organizations but is particularly important to those with big server farms, tight IT budgets, limited power supplies, or any combination of the above.

But this is only the first chapter of an unfolding server virtualization power-savings story. Organizations can realize even greater power savings by monitoring CPU utilization and/or CPU consumption at the physical server level and migrating virtual machine loads accordingly. These technologies are complementary in nature - and when used together provide significant additive value to the user environment.

Chapter Two
CPU utilization monitoring tools keep track of the utilization rates of physical servers within virtual environments, and when they fall below predetermined thresholds (e.g., below 30% utilization), automatically migrate virtual machines to other available servers. The actual movement of virtual machines among physical servers is typically done within the server virtualization platform, though it could be directed by a higher-level application. In either case, the movement should be dynamic and with little, if any, impact to performance. Should utilization rates increase and additional physical capacity be required to support virtual machines/applications, the process can be "reversed" and servers turned back on and "re-entered" into the physical server pool.

CPU consumption tools, in contrast, measure the actual power consumption at the node, or server platform level. In other words, they look at the actual efficiency of the physical servers and help organizations to ensure that power consumption is kept within targeted power budgets and, in the context of this discussion, that the power consumption, or efficiency, of each physical server is considered when migrating virtual machines on/off physical servers for maximum energy/power savings.

Emerging technologies exist today that enable this next "chapter" of power savings. While they haven't been rolled out in any significant fashion in production environments, they are already proving to have an important - and "additive" - effect on power reduction in test/dev environments in the data center.

Typically, servers in test/dev environments are only fully utilized during work hours. At night and on weekends, their loads are typically significantly less - often well below minimal server thresholds (i.e., below 30% utilization). Any are just idle. Yet, these servers remain powered on 24x7. Electricity is consumed to keep them running and, equally important, cool. By shutting these physical servers off during non-peak times, organizations can reduce power consumption costs significantly.

As an example, if four servers in a rack in a test/dev environment were kept running 24x7 and 16 were powered off (again, based on loads) 14 hours each day (during night hours), overall power consumption can be reduced by an additional 25% (beyond the savings gained from virtualizing in the first place). Figure 2 shows the potential savings progression in a 20-node environment from monitoring CPU utilization and migrating physical server loads accordingly and without. This chart does not calculate the potential cooling benefits resulting from powering down underutilized servers or the benefits of virtualizing the server environment in the first place (i.e., Chapter 1). Also, it does not consider the benefits from monitoring the actual power consumption at the node level. These will combine to driver server virtualization's power-savings benefits even higher going forward.

Conclusion
Just how much of a power-savings benefit an organization can expect to see from any one of these solutions (or combination thereof) will depend, not only on the cost of electricity ($ per kWh), IT utilization thresholds, server power, etc., but also which solution(s) they choose and how widely they are deployed (meaning how many applications it is applied to). But, there is no question that implementing server virtualization, on any scale, can help organizations reduce power and cooling requirements significantly. This is very important today, but going forward, it will be critical. Therefore, organizations need to consider the complete "power and cooling" impact when evaluating server virtualization platforms for the future.