Carbon emission trading schemes drive the use of data centre energy efficiency

Transcription

1 Carbon emission trading schemes drive the use of data centre energy efficiency Marc Demont April 19, 2011 Energy Economics & Policy Dr. Thomas Rutherford ETH Zurich

2 Table of Content 1 Introduction Data Center Data Center Components Power Utilization Indexes Swisscom Legacy Infrastructure Efficiency Cooling of IT Equipment Swisscom Best Practice Infrastructure Efficiency Emissions Avoided Kyoto Protocol Switzerland and the Kyoto Protocol Future CO 2 Regulations Models on CO 2 Emissions The Tax Model on CO 2 Emissions The Subsidy Model for Sustainability Emission Trading Emission Allowances for Companies The Efficient Level of Emission Swisscom Emission Allowances Conclusion References

3 1 Introduction Data centers use a significant amount of the nation's total supply of electricity. It is estimated that the total energy consumption of all servers and data centers in the U.S, in 2006, was about 61 billion kilowatt-hours (kwh), or 1.5 per cent of the total U.S. electricity consumption. The data center electricity usage has more than doubled between 2000 and 2006 and will double again by The energy costs in 2006 were about $ 4.5 billion ( [1]EPA 25). This consumption corresponds to the electricity use of 5.8 million U.S. households. The potential for improving the data center's energy efficiency is huge. Legacy data centers use between 35 and 50 per cent of electrical energy for cooling. A best-practice green data center can reduce cooling energy down to 15 per cent of the total. During the next years, carbon emission schemes can be introduced to data centers. These schemes are based on the electricity use of data centers and are then converted to CO2 emissions. A reduction of CO2 emissions means that a company that emits less than its allowance, could possibly sell unused credits to others. This paper shows the effect an introduction of carbon emission schemes can have on data centers. This is demonstrated by means of evaluating the data center of Swisscom, a large Swiss telecommunications company. The main goal is to reveal the benefits and incentives of emission trading. 2 Data Center A data center is a facility that contains all critical computing resources such as the mail, file, application and storage server and the operating system that runs them. It also includes the network infrastructure for routing. These resources run in a controlled environment and under a centralized management, allowing companies to operate around the clock. Applications range from internal financial and human resources to external e-commerce and business-tobusiness applications. Illustration 1: Inside a data center 2.1 Data Center Components A data center draws the electricity from the main supply electrical grid. Swisscom draws interruption-free, green energy from renewable sources. To ensure smooth operation of the IT systems, they must be free of interruptions. The electricity is distributed for the lighting, cooling and powering of the IT systems. Only a small percentage of the electricity is used for the powering of lights, the main part, however, is used to power and cool the IT systems. 3

4 Chillers Pumps Fans Servers Tranformer PDU UPS Storage Lighting Network Illustration 2: Data center components. Acronym keys: UPS - uninterruptible power supply. PDU - power distribution unit For the following description, see the illustration below. It shows all the power inefficiencies of a data center. In fact, less than half of the energy entering a data center actually reaches the IT systems. 1. This is the power drawn from the grid to operate the whole data center. 2. The power drawn from the grid passes through a power distribution unit (PDU) that transforms the power to the right phase and voltage. The power is also distributed to support systems (such as fire suppression, chillers, pumps and fans). As power is treated and transformed, some of it is wasted, due to inefficiencies. 3. The PDU runs the power through an uninterruptible power supply (UPS), which acts as a large battery. This part sets in, if an interruption in the power supply occurs. The UPS is dimensioned for a power supply of about 3 minutes before the diesel engine takes over. 4. Only a fraction of the electricity entering the data center is used to operate the IT systems. 5. As power is delivered to and consumed by IT equipment, one byproduct is heat. To ensure proper operations and prevent system failures, this heat must be removed by the same amount of cooling, which consumes even more energy. Cooling equipment typically involves fans, pumps, and chillers. Today, most systems produce and deliver cool air to IT and power delivery equipment by lowering the air temperature and by using fans to push the air to the equipment. 6. The produced heat is recycled or heats up the environment. 4

5 Illustration 3: Data center power inefficiencies 2.2 Power Utilization Indexes The power utilization effectiveness (PUE) index is a ratio of the total energy consumed by a data center to the energy consumed by the IT system itself. The energy consumed by the data center includes all the energy used for cooling, backup power, power conversions and lighting. The energy consumed by the IT system is the energy that flows from the output of the main supply to all the equipment racks. See no. 4 in the above illustration. PUE = total facility power / IT equipment power The ideal PUE rating would be 1.0. This means that 100% of the electrical energy delivered to the data center is used by the IT system. There is no power used to run the cooling equipment or the fans. If the IT system runs on an efficiency level of 80%, then 20% of the electrical energy will be dissipated as heat. Thus, the most efficient IT system does not generate any heat. A PUE rating between 2 and 3 is considered typical for a legacy data center. Current trends show that a PUE of 1.9 is considered readily obtainable, best practice solutions will yield a 1.3 index, while 1.2 is considered best practice. The inverse of the PUE equation is DCiE. The data center infrastructure efficiency (DCiE) is a nominal measure of the IT system efficiency. It is defined as DCiE = 1 / PUE = IT equipment power / total facility power x 100% 2.3 Swisscom Legacy Infrastructure Efficiency Server hardware is no longer the primary cost component of a data center. The purchase price of a new server has been exceeded by the capital cost of power and cooling infrastructure to support that server and will soon be exceeded by the lifetime energy costs for that server alone (EPA 30). A main goal of the Swisscom green IT strategy is to reduce energy costs in their data centers. This implies an improvement of the data center infrastructure efficiency (DCiE). To achieve a better efficiency, it is inevitable to improve the supporting components such as cooling, fans or lighting. A power reduction of the main IT system, however, is not possible. A major question is how to measure the IT system's power consumption in order to separate it from the power consumption of the entire data center. For the calculation of the PUE, "The Green Grid [6]" recommends 5

6 measuring IT system power consumption at the output of the uninterruptible power supply (UPS). See no. 3 in Illustration 3: Data center power inefficiencies. The following DCiE calculation represents the figures of the Swisscom data center in Zurich Herdern. Measured system Total facility power IT equipment power Uninterruptible power supply (UPS) Cold water production Cooling for air circulation Light, miscellaneous Power consumption in KW 2800 kw 1400 kw 280 kw 700 kw 336 kw 84 kw DCiE = 1 / PUE = IT equipment power / total facility power x 100% DCiE = 1400 kw / 2800 kw x 100 = 50 % PUE = 1 / DCiE = 2 Swisscom legacy infrastructure efficiency 12% 3% IT equipment power 25% 50% Uninterruptible Power Supply (UPS) Cold water production Cooling for air circulation 10% Light, miscellaneous Illustration 4: DCiE of a legacy Data center A power utilization effectiveness (PUE) that equals 2 is considered typical for a legacy data center. 2.4 Cooling of IT Equipment The purpose of a computer room air conditioning unit is to cool the servers and to enable applications to run. The unit itself, however, produces heat, which has to be taken away, thereby increasing energy costs. To improve the data center energy effectiveness, there are two determining factors which can be influential: 1. Improving the cold air circulation in the data center 2. Regulating the velocity of the air circulation Conventional cold air circulation in a data center is shown in the following illustration. The cold air circulates from the floor up to the roof of the facility. The fans of the IT components suck in the cold air and release it on 6

7 the other side. This is an inefficient way of air circulation because some of the cold air rises to the top without cooling any IT components. The heated air then passes through a water-cooled grid and back to the circulation system. The loss of energy efficiency in such a design is around 30% Illustration 5: conventional air circulation Best practice cold air circulation in a data center is shown in the following illustration. The cold air circulates from the floor directly into the rack where the IT components are. The fans of the IT components suck in the cold air and release it on the right side of the rack. In this set-up, no cold air is wasted into the atmosphere. In this case the loss of energy efficiency is less than 5% C C 16 / 22 C Luftgeschwindigkeit max. 2m/s Luftgeschwindigkeit max. 2m/s 16 / 22 C Leistungsabhängige Drehzahlregulierung Leistungsabhängige Drehzahlregulierung C C Illustration 6: best practice air circulation This second improvement highlights the regulation of the velocity of air circulation. 90% of all circulation air cooling units are still unregulated and running on full speed, irrespective of how much cooling the IT components need. The regulation in best practice systems measures the difference of the incoming and outgoing air temperature. Based on the result, the velocity is regulated. The following calculation presents the energy and cost savings achieved with a regulated air circulation. The measures represent the Swisscom data center Zurich Herdern. 7

8 Figures and Measurements Amount or consumption Quantity of air cooling units 14 Price of electricity Energy consumption of an unregulated air cooling unit Energy consumption of a regulated air cooling unit Difference regulated / unregulated CHF 0.24 / kwh 5.4 kw / year 0.26 kw / year 5.14 kw / year Energy savings per year = 5.14 kw x 14 units = 72 kw / year Energy cost savings = 72 kw/y x 0.24 CHF x 24h x 365d = 151,373 CHF 2.5 Swisscom Best Practice Infrastructure Efficiency With the described improvements, the results of the legacy and best practice data center design can be compared: Measured system Power consumption in kw Legacy System (per year) Total facility power 2800 kw 1945 kw IT equipment power 1400 kw 1400 kw Uninterruptible power supply (UPS) 280 kw 156 kw Cold water production 700 kw 292 kw Cooling for air circulation 336 kw 78 kw Light, miscellaneous 84 kw 19 kw Power consumption in kw Best practice system (per year) DCiE = 1 / PUE = IT equipment power / total facility power x 100% Legacy System DCiE = 1400 kw / 2800 kw x 100 = 50 % PUE = 1 / DCiE = 2 Best practice System DCiE = 1400 KW / 1945 KW x 100 = 72 % PUE = 1 / DCiE = 1.4 The following illustration compares both results graphically. 8

9 g CO2-Equivaltents per kwh Energy usage legacy (1,2) & best effort (3,4) in [KW] Light, miscellaneous Cooling for air circulation Cold water production Uninterruptible Power Supply (UPS) IT equipment power Illustration 7: DCiE index of legacy and best practice data center With IT equipment energy consumption of 1400 kw, the energy savings in comparison of DCiE 50% to 72% is Total facility power (old) Total facility power (improved) = 2'800 kw/y 1945 kw/y = 855 kw / year. The cost efficiency per year is: 855 kw x 0.24 CHF / kwh x 24h x 365d = 1,797,552 CHF 2.6 Emissions Avoided It is obvious that energies like coal or fossil fuels produce CO 2 Emissions. Also other energies like a nuclear power plant or hydroelectricity produce greenhouse gas emissions. These emissions result for example out of the electricity used for operating the plant. The following illustration shows the "grams CO 2 Equivalents per kwh" for several energy sources. greenhouse gas emissions over its entire life cycle nuclear hydroelectricity reservoir wood wind photovoltaics 0 Illustration 8: greenhouse gas emissions. Source: ETH ecoinvent

10 The following table shows the power consumption and the emissions avoided for the Swisscom data center. Since Swisscom draws green energy from renewable sources they use the following value for CO 2 Equivalents per kwh g CO 2 Equivalents per kwh Scenario Electricity Consumption (kwh) Old scenario Corbon Dioxide Emissions (tons CO 2 ) Improved operation kw x 365d x 24h = kwh 1945 kw x 365d x 24h = kwh Carbon Dioxide Emissions Avoided (tons CO 2 ) Through the improvement of the data center Swisscom reduces its carbon dioxide emissions by 110 tons. 3 Kyoto Protocol The Kyoto Protocol was negotiated in December 1997, in the city of Kyoto, Japan and came into force on 16 February "The Kyoto Protocol is a legally binding agreement under which industrialized countries will reduce their collective emissions of greenhouse gases by 5.2% compared to the year The goal is to lower overall emissions from six greenhouse gases - carbon dioxide, methane, nitrous oxide, sulphur hexafluoride, HFCs, and PFCs - calculated as an average over the five-year period of National targets range from 8% reductions for the European Union and some others to 7% for the US, 6% for Japan, 0% for Russia, and permitted increases of 8% for Australia and 10% for Iceland." (See Kyoto protocol [7]) 3.1 Switzerland and the Kyoto Protocol Under the Kyoto Protocol, Switzerland has committed itself to reduce its greenhouse gas emissions by 8% until 2012, compared to the 1990 baseline. As over 80% of Swiss greenhouse gas emissions account for CO 2, Switzerland has determined a CO 2 reduction target for this gas. This goal cannot be achieved with voluntary provisions only. Due to these facts, the Swiss government has decided on other provisions. Introduction of a CO 2 tax on fossil fuels Charge (0.01 CHF) on fossil fuels in the private sector ("Klimarappen") Tax incentives on biogenic fuels Since January 2008, a CO 2 tax has been levied on fossil fuels. The CO 2 tax is an incentive tax, which is redistributed to the public and the economy. It increases the cost of fossil fuels such as heating oil or natural gas. The higher prices are an incentive to reduce consumption and increase the use of clean energy. In the first year, 2008, the tax revenue was about CHF 220 million (at a rate of CHF 12 per ton CO 2 ). Since 2010, the rate per ton CO 2 has tripled to CHF 36. The tax levy was introduced because the targets in reducing emissions were not achieved. The tax revenue is redistributed evenly per capita. Hence, all those who benefit from this consume belowaverage amounts of fossil fuels. Companies who operate more sustainable will benefit from the CO 2 tax. The redistribution of the tax is coupled with the Swiss Old Age and Survivors' Insurance (AHV/AVS). Companies with many employees and low 10

11 energy usage benefit more from the redistribution. Companies with high energy consumption have an incentive to encourage energy saving measures. 3.2 Future CO 2 Regulations It is foreseeable that new regulations will have an effect the energy consumption for households and companies. Going back to the example of the Swisscom data center, the following models will show how the reduction of energy can influence the sustainability. 4 Models on CO 2 Emissions 4.1 The Tax Model on CO 2 Emissions The effect of a CHF 36 / ton tax is to shift the demand curve downward by CHF 36 at each quantity. The effects of such a tax are shown in the illustration below. Before tax, the price is PE. After tax, the price received by the seller is PS and the effective price paid by consumers is PC. Before tax, the quantity of energy sold is Q1; after tax the quantity decreases to Q2. P A S PC PE PS tax B D C E F D1 D2 Q2 Q1 Q Illustration 9: Tax model on CO 2 emissions A tax on CO 2 emissions reduces consumer surplus by the area B + C and producer surplus by the area D + E. The government collects tax revenue equal to the area B + D. As mentioned above the revenue is redistributed evenly per capita. Not mentioned above is the deadweight loss when a tax distorts a market. The area C + E measures the size of the deadweight loss. Hence the total surplus in the market falls by the area D + E. An efficient tax system tries to minimize the costs to taxpayers and the government. The above figure doesn't show the government costs to manage the redistribution of the tax revenue. 4.2 The Subsidy Model for Sustainability The effect of a subsidy is to shift the demand curve upwards. The effects of such a subsidy are shown in the illustration below. Q represents the quantity of sustainable technologies in demand. Before the subsidy, the price is PE. After the subsidy, the price received by the seller is PS and the effective price paid by consumers is PC. Before the subsidy, the quantity sold is Q1; after the subsidy the quantity increases to Q2. 11

12 P Supply PS PE subsidy PC D2 D1 Q1 Q2 Q Illustration 10: Subsidy model for sustainability 5 Emission Trading Emission trading is the most important instrument in the Kyoto protocol. The basic goal in trading emissions is to limit the permissible total and break them down into tradable units. Thus, the before unlimited greenhouse gas emissions are shortened and become a price for trading. The Kyoto protocol provides that an international trade of emissions between industrialized countries will exist. The countries will receive a limited amount of emission allowances from the UNO. One emission allowance stands for 1 ton of CO 2 equivalents. For every ton CO 2 equivalent emitted the country has to give back one emission allowance. Surplus allowances can be sold to other countries. If a country doesn't achieve the reduction target, it has to purchase additional emissions allowances. Through the trade of emission allowances the greenhouse gas emissions will by reduced by the countries where it is more cost-efficient. If a country is in the position to reduce its emissions it will do so and sell the excess allowances. 5.1 Emission Allowances for Companies In Switzerland the government allocates the emissions allowances to the individual companies. This enables the Swiss economy to profit from the allowed flexibility of the Kyoto protocol. To fulfil the reduction target, the companies now have three possible strategies to adopt. Reduce the emissions through own actions Purchase additional emissions allowances Compensation of emissions through investment in emission reducing projects abroad. 12

13 5.2 The Efficient Level of Emission An efficient level of emissions is one that maximises the net benefits from pollution respectively minimizes the sum of total abatement costs plus marginal social costs. $ / unit of emissions MCA MSC E0 E* E1 Level of Emissions Illustration 11: Efficient level of emission The MSC curve shows the marginal social cost and the MCA curve shows the marginal costs of abatement. At E 0 the marginal cost of abating emissions is greater than the marginal social cost. At E 1 the marginal social cost is greater than the marginal cost of abatement The efficient level of emission is E* where MCA = MSC. 5.3 Swisscom Emission Allowances As shown in chapter 2.6 Emissions Avoided Swisscom avoids tons of CO 2 through its best practice solution. In the following example we assume that Swisscom avoids 100 tons of CO 2 in total. With these figures, Swisscom could trade 50 emission allowances to an other company. Start Some company Swisscom Previous CO 2 Emissions: 1000 tons Previous CO 2 Emissions: 1000 tons CO 2 Reduction Emission Allowance: 950 tons Emission Allowance: 950 tons Actual CO 2 Emissions: 1000 tons CO 2 Emissions: 900 tons Trade Additional Purchase: 50 tons Sell: 50 tons 13

14 Emissions trading can be more beneficial for both the buyer and the seller. With the assumption from the above example we assume that Swisscom can abate its CO 2 at a cheaper cost than company B. For Swisscom the marginal cost of abatement (MCA) is low and on the other hand, company B has a higher MCA. Swisscom has an elastic MCA curve where the curve of company B is rather inelastic. The value R Req represents the total amount of emissions that need to be reduced by the companies. At R* the market allowance price (P) intersects the MCA curve. This is the efficient level of emission abatement. [10] MCA ($ / unit) Swisscom MCA ($ / unit) Company B MCA E MCA P A B P D F C RReq R* Emissions abated R* RReq Emissions abated Illustration 12: Emission Trading (Cap and Trade Model) On the Swisscom side, at R Req the MCA curve has not intersected the market allowance price (P). This implies that Swisscom has potential to abate more emissions than required. On the company B side, the MCA curve already intersects the market allowance price (P) at R*. The required emission abated lies above the market allowance price. Thus, company B saves costs if it abates less emission than required. Swisscom abated more emissions than required (R Req ). Through that additional abatement, Swisscom is able to make a profit which is shown in the triangle A, B, C. Its surplus (R* - R Req ), Swisscom can sell the additional credits to the company B and is paid the price P for every unit it abated. Total revenue is the area R* - R Req x P. The company B buys emission allowances from Swisscom for the price P. The amount needed is R Req R*. Through that the company B is able to save costs equal to the triangle D, E, F. In this case of trade both parties are better off. 6 Conclusion The trend in IT shows that the performance need for servers, databases and network components increases every year. This has to do with increasing internet traffic and the need for more storage capacity. This implies more servers or more powerful hardware. This again means higher energy consumption. The past experience has shown that the necessary energy in Swisscom data centers is limited. It is becoming increasingly important to improve the energy efficiency in data centers. The above measures about the data center are describing just one area of the data center Zurich Herdern. To project the measures to all the existing 14

15 Swisscom data centers, a tremendous energy efficiency level could be achieved. These figures also show that the price of hardware plays a comparatively small role in contrast to the energy costs. For future capacity planning it is unavoidable to have an energy measurement dash board. These measures provide financial and technical data for accounting and planning. As a baseline, the today's measures will be used for future decision taking and planning. This allows ongoing comparisons. Until now only the Swisscom datacenter Herdern is equipped with such a measuring system. The flexibility of the Kyoto protocol gives the companies an incentive to invest in green data centers. An abatement in energy leads to a surplus in emission allowances, which can be sold to other companies. The gained profit can be invested again. 15

16 7 References [1] Environmental Protection Agency (EPA) [2] Gartner (8 October 2008): Green Data Centers: The Six Key Attributes of Data Center Energy Efficiency Metrics [3] Gartner (29 June 2010): Market Trends: Sustainability and Cost Savings Objectives for Large Data Centers Will Require Tighter Efficiency Metrics [4] Data Center Energy Forecast (July 29, 2008, Silicon Valley Leadership Group [5] EPA Data center Report Congress [6] The green grid: White Paper 22 PUE DCiE Usage Guidelinesfinalv21.pdf [7] [8] Emission Trading "Ein markwirtschaftliches Instrument im Klimaschutz" [9] Emission Trade in Switzerland: [10] The Real Climate Debate: To Cap or To Tax? 16