The idea of the cloud is based on IT as a utility;plug in and get the resources you need ÔÇô readily available and flexible. Virtualization is a tool to get there. As more businesses look to the cloud, their focus is on effective virtualization strategies.

The ability to consolidate a number of workloads onto one machine is the most disruptive force driving data center operations today. Given the potential benefits, pilots are easy to justify. With initial success, these pilots quickly lead to full projects. Our modeling suggests that as IT professionals use virtualization and consolidation to manage system utilization, improvements in data center energy efficiency are guaranteed.

As with most other disruptive forces, however, the "Law of Unintended Consequences" holds. While adoption of virtualization technologies shows clear benefits, counter-intuitive results sometimes mislead data center managers. Data center efficiency metrics may move in the wrong direction;organizations may miss opportunities for additional improvement;IT may become too reliant on these new technologies for day-to-day management.

UNINTENDED CONSEQUENCE 1: PUE CAN BE VIRTUALIZATION'S FIRST VICTIM
The first unintended consequence is the effect of a virtualization program on a data center's Power Usage Effectiveness. PUE specifies how efficiently a data center distributes utility power to IT equipment. It provides guidance for power distribution and cooling architectures but not the deployment or operation of IT equipment. Changes to the IT environment can move PUE in the wrong direction, even though the power and cooling architectures have not changed.

Metrics like Energy Intensity may provide a better indicator of the effects of virtualization on power consumption. More commonly found in economics texts, the expression relates an activity's resource consumption to its result ÔÇô any change in this metric is directly realized in a corresponding energy-consumption change. With this approach, the true magnitude of power benefits becomes clear.

UNINTENDED CONSEQUENCE 2: MISPLACED GAINS
The second unintended consequence is the effect of consolidation on data center space utilization. Decommissioning of servers recovers power, cooling and rack-space capacity. A virtualization/consolidation program that does not consider location can create isolated holes in the data center instead of larger, more useful, islands of whitespace. Two or four empty U's in a rack allow only for deployment of another server. An empty rack or a whole row enables a wider range of options. The most effective consolidation programs incorporate plans for creating and re-purposing whitespace. This requires IT organizations to bring facilities teams into the fold. This is also where a focus on server age pays off. Occasionally, older machines are excluded as consolidation targets because their lower compute potential and smaller memory capacity make them appear to be better utilized. These servers cluster in the data center's oldest parts, areas most likely to present the greatest improvement opportunities. Strong consolidation programs adjust for this by ensuring that the oldest servers deployed in their data centers are included as targets regardless of performance statistics whenever possible.

UNINTENDED CONSEQUENCE 3: UNDERESTIMATED BENEFIT TIMEFRAME
The third unintended consequence is more subtle: data center management strategies that assume future power and cooling concerns addressed by the virtualization platform will be minimally effective until 85-90 percent of the data center's work is virtualized. Any virtualization program past a pilot will divide the data center into two classes of equipment: resource hogs and efficient machines. Servers still running standalone applications will be the hogs, remaining the primary energy and space consumers. Servers running virtual workloads will be the efficient machines. Consider a data center, pre-consolidation, that manages 1000 workloads on 1000 servers. With a consolidation program aiming to consolidate half of those workloads onto virtual machines at a 10:1 ratio, the data center will end up with roughly 50 physical servers acting as hosts for virtual machines, but still have 500 physical servers running standalone applications.

Virtual machine hosts may use more power per server than machines running standalone applications but Energy Intensity of this work will be much lower given the number of workloads they run. This is all while total energy consumption has dramatically decreased. Eighty to ninety percent of total data center power consumption will still be due to un-virtualized workloads and not directly addressable by the virtualization platform's management tools. For the time being, data centers are going to continue supporting some standalone workloads. IT managers must make appropriate configuration decisions. A server running a standalone application exhibits lowest Energy Intensity when it is right-sized for its application and includes/uses the latest energy-efficient features.

Servers running many virtual machines are often heavier configurations with greater average power consumption, presenting an opportunity and an issue. Mixing these servers with energy-efficient servers running standalone applications in one rack can lower average rack power densities, potentially avoiding the need to accommodate significant increases in power and cooling provisioning.

As in any complex system, changing one element radically can have unexpected results. As we learn more about effects of these projects, the unexpected becomes foreseeable and addressable. For virtualization/consolidation projects to reach their full potential, they should incorporate strong data center metrics with an understanding of the full relationship between compute workloads and physical equipment, but also include a clear vision for the entire data center: facilities, IT, equipment, workloads and management.