A Grid-Wide Strategy For Sustainable Data Centers
A mutually beneficial solution will solve tomorrow’s problems now.
Electric consumption in the U.S. has remained flat the last five years. During that same period, device connectivity has proliferated, media streaming has become ubiquitous, and data center electric consumption has nearly doubled to over 3% of demand. To keep costs low, data centers are typically located in areas with cheap electricity on grids dominated by coal-fired power plants. Concerns of electric price inflation and increasing pressure to reduce their carbon footprint are causing a growing number of IT companies to sign RE power purchase agreements (PPAs). With the price of electricity from these long-term agreements approaching grid parity, they not only provide low-cost insurance against price shocks, but also earn them kudos for “greening the-grid.”
The predicament facing utilities is different. Demand is shrinking, they face new regulatory challenges, and they are generally perceived as hostile towards renewables. Despite the flat demand growth, federal and state mandates direct them to invest in expensive new cleaner capacity, while retiring their fully depreciated assets. To satisfy state renewable portfolio standards (RPSs), this new capacity often includes renewable wind and solar generation. With the rapid growth of residential solar, their revenue base is being further eroded, and a larger share of future costs will be supported by a fewer number of customers. As witnessed in Germany, the reliability of the grid will decline and electric rates will almost certainly increase.
Resolving their problems in isolation, utilities and IT companies pay little regard to the impacts they have on each other’s business. IT companies are embracing change while utilities resist. IT companies are content with low-cost grid electricity and GHG credits while utilities fight encroaching technologies through a combination of rate impact studies and regulations. Both sides are missing an opportunity. IT companies are neglecting to consider the future grid impacts, and utilities are failing to see how data centers can help resolve their current dilemma.
In 2007, the EPA presented a report to Congress, documenting the doubling of data center electric consumption between the years 2000 and 2005. They further testified that over the following five years consumption would again double, placing increasing demands on already congested grids. In 2011, Koomey published his analysis, which contradicted their findings, estimating that the actual increase was only 36%. The growth rate of energy consumption was seen to be tapering, fears were calmed, and the impetus to do anything other than efficiency improvements was lost.
Despite the recession and anemic recovery, the electric consumption of data centers has continued to grow. Internet connected device proliferation, a growing propensity for media streaming, and the emergence of cloud computing, are all demanding increased network capacity. Even utilities, as they evolve to a digitally controlled grid, are adding to the problem, not to mention the growing interest of large corporations in Big Data. This near constant connectivity and growing data dependency is causing cloud traffic to increase at a compound annual growth rate of 32%.
As represented in Figure 1, both data centers and utilities are facing pressures to lower their carbon footprint, increase efficiency, and reduce costs, all while maintaining or improving reliability and security. Although they share very different outlooks, and their current paths appear in conflict, both utilities and IT companies would benefit from cooperation. Embracing a common strategy and working together they could identify mutually beneficial solutions and resolve the many challenges their common problems impose.
Although the EPA miscalculated the increase in data center energy consumption, somewhat discrediting the report, they otherwise provided a comprehensive analysis of how data centers impact the grid. In particular, they recognized that they are frequently clustered in transmission congested areas, raising awareness of the increased risk of decreased energy security. This tendency for clustering, which remains dominant, is now influenced by other factors. No longer only concerned about the current cost of electricity, IT companies now worry about the expected future costs.
Estimating the rate change is not easy. Besides the fuel cost assumptions, calculations are complicated by generation mix, planned and potential retirements, and the progress towards RPS targets. To hedge against these uncertainties some of the larger companies now favor long term PPAs from renewable sources. With the price of wind and solar falling, becoming closer to grid parity, the ability to secure them is now often a prerequisite. Providing fixed low cost energy and renewable energy credits (RECs), they are becoming a primary vehicle for renewable deployment. This recent trend is increasingly placing data centers alongside renewables, and connected to grids with locally high penetrations of renewable energy.
Renewable energy is inconsistent and susceptible to sudden load swings. To counter this uncertainty, and overcome issues from increased penetration, utilities and system operators are exploring new solutions. Recent innovations in generators and customer programs offer some help, but many experts believe that eventually energy storage will also be required. Absorbing energy when generated in excess, these energy storage devices release it at times of deficiency through a process called load shifting. Pumped hydro is the most mature and cost effective utility scale technology, but due to capacity limitations, is not viable. While a breakthrough in other options has so far remained elusive, much hope is placed on utility scale battery technology.
BALANCING THE GRID
Utilities and system operators are responsible for maintaining the equilibrium between electric supply and customer demand. They achieve it using spinning reserves, demand response (DR), and for extreme events, fast starting peaker plants.
In markets where the ratio of peak to average consumption is high, the low availability of spinning reserves already causes difficulty. To overcome this, and the added complications from increased RE, much attention is now focused on speed of response. Encompassing demand and capacity management technologies, responsiveness is increasingly associated with new solutions and innovations for balancing the grid.
Electricity demand, as shown in Figure 2, is constantly changing as the result of the residential, commercial, and industrial customer loads. These fluctuations normally follow reasonably predictable patterns and are easily managed. Complications are introduced by unplanned maintenance, and seasonal events that can drive unexpected heating or cooling loads. Under these circumstances grid management is more challenging. When renewables are included, the greater uncertainty causes additional complications, increasing grid management problems, and adding to the likelihood of unscheduled generator dispatch.
The electric output from renewables cannot be scheduled. Figure 3 shows an example wind output and corresponding load demand. For this period, the wind output varied from near 0% to 20% of the demand, providing the least support when most needed. As the RE penetration increases, reserves to cover its variability also increase, dragging down efficiency, and impacting GHG reductions. This problem is not limited to the U.S., it’s worldwide. To address it, experts are seeking practical and economic solutions for balancing the grid. They have identified system flexibility, as one of the most essential elements of a robust infrastructure.
Flexible technologies fall into three categories, supply, demand, and storage. Both supply and demand technologies are maturing, but utility-scale storage, proclaimed by many as the most crucial, is still only developmental. While lead acid batteries are functionally suitable, their application to grid infrastructure is limited by safety and environmental concerns. The other options are expensive, geographically constrained, or unproven. Despite this lack of success, expectations remain high, and are bolstered by the related efforts of automotive companies.
Several companies offer flexible generation. Combustion turbines and natural gas combined cycle power plants (NGCC) are the dominant solutions. Together they account for 50% of recent utility scale additions. Modern NGCC systems start quickly and ramp load aggressively, providing similar functionality to the less efficient combustion turbines used only for peaking. Historically, influenced only by demand, NGCC power plants have operated at a capacity factor of 45%. The less efficient plants with poor flexibility run close to base load. Low utilization, and the increased maintenance from cycling, increases the cost of electricity. Therefore it is reasonable to expect even higher future costs as RE gains momentum, and the grid requires more severe ramping and lower utilization from their fleet of NGCC plants.
DR programs offer a mechanism for shedding load. Using these programs, a utility can prioritize the shutdown of customer loads, thereby curtailing peak demand. Successful implementation, as shown in Figure 4, significantly reduces the frequency of these critical dispatch events. Being a low cost and instantaneous solution, DR has become a key element of grid management. The Federal Energy Regulation Committee (FERC), a strong proponent of DR, expects adoption will continue to gain acceptance, estimating it will grow to around 55,000MW, or 4.5% of the peak resource portfolio, by 2021.
To counter wind and solar fluctuations, the grid needs flexibility. It requires the ability to respond quickly to both excess and insufficient capacity. DR only addresses shortfalls. Overcapacity, never considered a systemic problem before RE, is a relatively new concern. To resolve power surges from renewables, fast ramping NGCC plants are being deployed. While they rapidly add or shed load efficiently, their electricity is more costly, and it will rise with increased RE deployment. For the grid to work effectively, without inflating electric prices, there is a vital need for alternatives. Solutions that will allow these plants to run at higher utilization, suffer less maintenance, and generate electricity at a lower cost.
Data centers depend on the electric grid for their power. To achieve the necessary level of reliability they typically use dual electric services fed from independent grids with UPSs and emergency generators for backup. This has been the standard practice for many years. Today, some of the larger IT companies are demonstrating their commitment to RE by installing fuel cells, solar PV, and wind turbines. The Apple facility in Maiden, NC, for example, currently has 4.8 MW of fuel cells, and 20 MW of solar PV, with plans to increase the outputs to 10 MW and 40 MW, respectively.
Like Apple, many larger corporations have initiated GHG reduction programs and are consequently investing in renewables. To reconcile their earlier decisions, they applied escalation factors, forecasting that grid electricity would rise, while RE costs would decrease. Today, more IT companies are investing in renewables. Some invest because of the environmental concerns, but most because of customer pressure and the fear of electric price inflation. Whatever the reason, these companies with large appetites for energy are driving RE deployment and increasing its competitiveness with the grid.
Fuel cells have received much fanfare for data center onsite energy; however, their high cost remains a major deterrent to wider adoption. They are one of the most expensive forms of power generation. Bloom Energy’s fuel cells cost over $8,000/kW, and viable only after incentives. Although other fuel cells are half the price, none have rivaled the success of Bloom. Whatever the selection criteria, without the continued assistance of subsidies, which expire in 2016, fuel cells will be relegated to “niche markets,” where purchase decisions are based on factors other than return on investment.
An improved onsite energy solution would include clean, competitively priced generation that can smooth out the fluctuations from local wind and solar. It could operate in several modes, as outlined in Table 1, and also function as backup generation. More importantly, it could be fully employed in grid stabilization programs. In 2011, PJM paid over $173,000/MW for demand response capacity. At a generator cost of $500/kW, this corresponds to a payback of less than three years, and for new installations, offset by the cost of traditional backup, they would be revenue positive immediately.
Data centers with large power demands are drawn to locations that offer low cost electricity. This places them on electric grids dominated by large coal-fired power plants. To comply with the new emission rules, many of the older coal plants are being retired. They are being replaced with RE or plants fueled by cleaner burning natural gas. These actions increase the cost of the generating fleet and raise the price of electricity. The circumstances facing utilities are exacerbated by the lack of demand growth and the acceleration of residential solar. To overcome these issues, utilities need to evolve their business models, and the grid needs innovative technologies and concepts to accommodate the rapidly changing generation mix.
IT companies are aggressively pursuing opportunities to protect their profits. To hedge against energy inflation they are purchasing long-term carbon free PPAs. This strategy supports both the fiscal responsibility demanded by their shareholders, and the environmental stewardship valued by their customers. This migration towards greener energy is increasing the penetration of renewables; however, it’s also reducing the reliability of the electric grid. Unless viable solutions to counter their disruptions are developed, this near-term positioning by IT companies will eventually undermine the foundation of reliability on which data centers are built.
Although many studies have been performed, there is no consensus at what level of renewable penetration the grid becomes unstable. The experts do agree, however, that to support its growth, we need greater system flexibility. While both demand and generation solutions are commercially viable, there is a growing preoccupation that energy storage holds the key. The most favored solutions involve new battery chemistries. Small batteries for portable devices are commonplace, however, larger utility-scale solutions are elusive, and remain costly, unproven, and inefficient.
By incorporating clean flexible generation in place of diesel gensets, data centers, can provide similar functionality to energy storage. Accepting power when the grid has a surplus, and self-generating at times of high demand. They can also balance the output from local renewables. Utilities would benefit from the large demand or dispatch capacity. Data centers, through revenues from load balancing services, would see lower electric bills. Grid integrated data centers, strategically located throughout the electric network, offer a near-term, low cost solution to grid instability.
What appears to be an adversarial relationship could actually be symbiotic: the carbon intense power generation industry of the past, working with the green IT companies of the future to solve each other’s problems; wind and solar farms, built close to data centers with flexible generation, reducing transmission congestion and providing dispatchable assets to stabilize the grid; dynamic data centers reducing the problems from increased RE penetration, providing the functionality to not just protect the data center, but to facilitate continued RE growth, while preserving the efficiency and reliability of the grid.
IT companies are looking for ways to lower their operating cost and reinforce their commitment to green energy. Utilities are trying to fulfill their GHG obligations, maintain grid stability, and contain costs. Data centers, with flexible onsite power, can convert a once benign load, into a multi-functional dynamic resource. They could become an intimate part of the grid, not only assisting RE integration, but also securing continued access to low cost reliable grid electricity.