Services & Maintenance: Data Center Energy Efficiency

Contributor Steve Yellen writes about how the interconnection of facility and technology concerns in this space require a comprehensive approach.
Contributor Steve Yellen writes about how the interconnection of facility and technology concerns in this space require a comprehensive approach.

Services & Maintenance: Data Center Energy Efficiency

Services & Maintenance: Data Center Energy Efficiency

By Steve Yellen
Published in the February 2009 issue of Today’s Facility Manager

In one variation of the famous story, “The Blind Men and the Elephant,” four blind men were asked to identify an elephant by touching only one part of it. One man said it felt like a pillar; another said it was a rope. The third blind man thought it was a tree branch, and another, still, said it was a pipe. They were all right—and also wrong. Because they only looked at a part of it, they couldn’t identify the whole. Their guesses depended on whether they felt its leg, tail, trunk, or tusk.

For facility managers (fms) in many organizations, the elephant in the room is the lack of a holistic management plan for the data center. They can only see part of the problem. Typically, the facility management (FM) department might own everything up to the rack, including providing power to the rack itself. The IT team picks it up from there, taking responsibility for everything that’s in the rack. Every facility site is different, but the shared ownership and interdependence between FM and IT are common to nearly all of them.

This makes it difficult to meet today’s challenges, particularly when it comes to the task of reducing power bills. Between 2002 and 2006, global electricity prices rose by 56%. At the same time, the demand for computing capacity ballooned in line with greater business demand for IT services. The end result is an annual power bill for all U.S. data centers that has been estimated as high as $3.3 billion. As the recession looms, fms will come under increasing pressure to meet or reduce this budget.

Meanwhile, data center managers are tasked with making their existing resources go further. In a survey by the Data Center Users Group (DCUG), a group of data center managers, IT managers, and fms formed by Emerson Network Power, 46% of respondents cited said the amount of power available as the top constraint on growth in their data centers. If data centers and the supporting infrastructure can use less power, organizations can unlock lost capacity while also saving money.

Pursuing A Metric
How can improvements in energy consumption be tracked? The Green Grid, an industry group focused on data center efficiency, has proposed Power Usage Effectiveness (PUE) as a way to measure the energy efficiency of a data center. The PUE is calculated by dividing the amount of power entering a data center by the power used to run the computing infrastructure within it.

According to The Green Grid, the PUE figure can theoretically range from one up to infinity, with a value of 1.0 indicating that all power was being used by IT equipment with 100% efficiency. To take an example, if the equipment in a data center consumed 500kW, and 750kW was being supplied to the facility, the PUE would be 1.5.

The PUE is a useful yardstick, but there is a risk that it is applied too broadly. The metric is a good way to benchmark how well the power and cooling are right sized to match the IT load. However, it doesn’t reveal anything about what IT is doing within the data center, which can lead to some counterintuitive results.

For example, perhaps the IT team uses virtualization (a technique where one physical server is partitioned into multiple virtual servers which operate independently). That could enable the department to eliminate half its physical servers, and the power consumed by IT would drop dramatically. Any savings in cooling would be relatively modest, however. The result is that power would be saved, money would be saved, but the PUE would be worse than it was before. It would look like the facilities team was consuming too much power for the IT load, if the PUE was studied in isolation.

Many fms have been tasked with delivering a good PUE in the interests of cutting back power bills. However, it could be argued that this is like trying to clean an elephant caked in mud by washing its tusks; it can result in very shiny tusks, but very muddy footprints. PUE is a good measure of infrastructure efficiency, but a poor indicator of the efficiency of the total operations. If the IT load is inefficient, it is still possible to have a good PUE by right sizing the power supplied to match that inefficient load.

In order to address the energy challenge, IT and FM teams must collaborate. Having a metric they can share and that can reflect their individual and joint efforts to save power could help them achieve their organizational goals. Such a metric should drive the right behavior and enable organizations to focus on data output, energy efficiency, or both. It ultimately needs to be available and published at the IT device level in order to drive the right buying behavior for new equipment.

Most importantly, the metric agreed upon needs to scale from the device up to the whole data center. That’s because there’s a cascade effect where changes at the device level can result in considerable savings at the facilities level. Every watt saved at the server component level (such as processor, memory, hard disk) results in an additional 1.84 watt savings in the system—the power supply, power distribution system, UPS system, cooling system, and building entrance switchgear and medium voltage transformer (see Figure 1).

Figure 1. In evaluating energy usage in data center facilities, fms should consider a cascade effect. Changes at the device level can result in considerable savings at the facilities level. Credit: Emerson Network Power.

There has been another metric proposed to measure energy efficiency in data centers—compute units per second per watt (CUPS/W). This metric takes account of improved IT performance as well as gains in energy efficiency.

The U.S. Environmental Protection Agency (EPA) estimates that data centers around the nation will increase their total electricity consumption from 61 billion kilowatts in 2006 to 100 billion kilowatts by 2011. That number only tells half the story, though. Just as important as energy efficiency will be computing efficiency—how much processing work that electricity buys at 2011 prices. The CUPS/W metric calculates that factor.

Just as the miles per gallon benchmark enables car buyers to decide which is the most fuel efficient model, an industry wide, universally agreed upon metric would enable data center managers and fms to compare the overall efficiency of these spaces and focus on tactics that can improve underperforming centers. For example, managers at an organization operating two data centers with different results could try to identify why one center is operating less efficiently than the other.

The metric could also be posted on equipment, similar to how Energy Star ratings are used. So far, there is no standard metric that the industry has agreed upon, but CUPS/W is helping to stimulate debate about what such a metric could be.

Outside Forces
Addressing energy efficiency cannot be solved entirely within the data center space. It requires a change in organizational culture too. The number one reason power usage is increasing in data centers is that demand from “the business” is rising. Making business managers accountable for the energy their applications consume could have significant impact on an organization’s energy use.

One of the best practice approaches that can achieve this is to charge business departments for the energy consumed by the IT they commission. In a recent study, the Aperture Research Institute found that only 24% of data centers do this. Many organizations will be thwarted by a lack of infrastructure in any attempt to charge business departments for the energy consumed by their IT. Only 40% of those surveyed said they monitored server power consumption. A further 44% said they were considering introducing it, but 16% confessed they had no plans to monitor servers.

It is impossible to charge the business for using a resource that cannot be measured. And, until power consumption is accurately measured, data centers will be unable to charge business users accurately, making it difficult to link provisioning decisions to their energy consequences.

Putting Heads Together
Although the FM team might be tasked with saving electricity in its data center operations, it cannot do that alone. The team needs to cooperate closely with the IT managers and agree on measurements that will encourage optimal behavior across teams. It also needs to win the support of the wider business, in particular with regard to requests for IT services. That could be achieved by charging business divisions for the total cost of IT services, including power.

As the elephant story shows, it’s impossible to get a clear view by focusing on isolated details, which could arguably include the PUE. Through a holistic management strategy, however, fms can be empowered.

Yellen is the vice president of product and market strategy for Aperture Technologies, a Stamford, CT-division of Emerson Network Power.

Suggested Links:

You Might Like: