Cloud computing's often portrayedas a bastion of efficiency, replacing bloated legacy data centers with a flexible new architecture that saves time, equipment, maintenance and energy.
But that real picture is far more complex, and deployments can actually have adverse energy effects if not managed properly.
Some power drains are easy to spot and fix, such as leftover legacy servers that haven't been fully retired and are still drawing power, or poor due diligence that results in choosing cloud providers with inefficient data centers. But other threats are harder to identify, including poor server utilization resulting from inefficient end-user provisioning of servers; and servers and hardware that are part of the cloud strategy and relatively new, but not new enough to efficiently handle increased loads once a high level of utilization is achieved.
Paul Stemmler, a managing director for North American Infrastructure at Citigroup, says the environment created by cloud computing can contribute to server "sprawl." "You have to watch the server utilization rates, and look to reclaim underutilized assets."
Sprawl reduces energy efficiency because users provisioning individual programs that are easily accessed via the cloud may lay claim to more server space than is actually needed.
"If something's cheaper and easier to order, people will likely consume more," Stemmler says. "And without appropriate controls and governance, this can lead to higher energy use."
Bernard Golden, CEO of cloud computing consultant Hyperstratus, says overall efficiency is correlated with utilization-servers use about the same energy to operate and cool themselves regardless of utilization. "If the power used in computing is low [which happens at low utilization], the [power utilization] ratio will be bad," he says.
Stemmler says Citi's removing about 4,100 servers per year through cloud and virtualization, yielding an additional savings of about $1 million in energy costs per year. To help achieve those savings, the bank closely monitors utilization rates of its servers, which enables the institution to locate and "reclaim" underutilized assets.
Tim Elkins, evp and CIO of PrimeLending, a Dallas-based residential mortgage lender that's in the midst of a private cloud deployment, uses a feature of VMware called Distributed Resource Scheduler to achieve the best utilization possible. Also of concern is hardware involved in the cloud, since some equipment often can't efficiently operate at a higher utilization rate. "You need to understand how the current environment is performing and review the proposed architecture to determine whether the servers that are currently in existence can be repurposed to handle a cloud structure," says Anthony Rydell, a senior manager at Deloitte Consulting.
Golden says Intel and AMD have made server energy efficiency a priority over the past five years, and are designing their chips to better support virtualization.