Posted by Todd Boucher
on 08/2/2010
Virtualization technologies today are allowing customers to consolidate servers at ratios of 15:1, 20:1, and possibly more, helping IT teams increase operational efficiency and decrease rack footprint requirements inside their data center. In addition, organizations are leveraging virtualization for the energy improvements it provides them, as the total decrease in energy realized by consolidating multiple legacy “volume” servers far outweighs the energy consumed by one more high-powered blade center.
To encourage these efficient practices in the data center, we are starting to see utility companies provide incentives for virtualization/consolidation projects. Although not universally offered, we have seen existing prescriptive programs where utilities are offering incentives of $100, $200, and even $250 per server removed or consolidated. By making these programs prescriptive, utilities are essentially equating the replacement of a volume server with a virtual server (through consolidation) to the replacement of a T12 lighting fixture with a HPT8. In other words, virtualize your servers and you are automatically more efficient.
Or are you?
What these programs fail to consider is the implications of virtualization on the power and cooling infrastructure in the data center. A typical data center with a raised floor and perimeter downflow CRAC units would traditionally support 1-3kW per rack of IT load. Before virtualization, these densities could easily be supported. However, through the use of blade server technologies often implemented during virtualization projects, rack densities are increasing exponentially to 10-20kW per rack (and more). Thus, cooling infrastructure designed for 1-3kW per rack is forced into futile attempts to keep up.
The resulting hot spots created by the 20kW rack loads leave our customers wondering how they could still be facing cooling shortcomings after significantly decreasing total IT load in the data center. Inevitably, they realize they are not able to physically cool a 20kW rack through a raised floor and resort to supplementing their HVAC system in an attempt to control the heat load—adding spot coolers, decreasing Entering Water Temperature, decreasing Supply Air Temperature—are some of the many examples we have seen customers take to combat hot spots. One customer we recently worked with, after successfully consolidating servers at a 15:1 ratio, added 15 tons of cooling to the already oversized HVAC infrastructure that suddenly could not keep up.
The net impact on energy consumption is clear. Although modest reductions in energy consumption were made through server consolidation, the resulting 15 ton (52.5kW) cooling addition created a net increase in energy consumption from the point at which the customer started the consolidation project. In other words, their virtualization project increased their energy consumption in the data center.
We commend the utility companies’ awareness of energy consumption in the data center and their efforts to incent customers to become more efficient. However, it is important to look at energy consumption in the data center holistically and understand the behavior that occurs as a result of these incentive programs. Virtualization will affect each customer’s environment differently, but by incenting customers to virtualize without first educating them on related power and cooling requirements, are we truly becoming more energy efficient?
