Energy Efficiency Improvements Using DC in Data Centres

(1)

Energy Efficiency Improvements

Using DC in Data Centres

Sofia Bergqvist

(2)

Abstract

The installed power usage in a data centre will often amount to several megawatts (MW). The total power consumption of the data centres in the world is comparable to that of the air traffic. The high energy costs and carbon dioxide emissions resulting from the operation of a data centre call for alternative, more efficient, solutions for the power supply design. One proposed solution to decrease the energy usage is to use a direct current power supply (DC UPS) for all the servers in the data centre and thereby reduce the number of conversions between AC and DC.

The aim with this thesis was to determine whether such a DC solution brings reduced power consumption compared to a traditional setup and, if this is the case, how big the savings are.

Measurements were carried out on different equipment and thereafter the power consumption was calculated. The conclusion was that up to 25 % in electricity use can be saved when using a DC power supply compared to the traditional design.

Other benefits that come with the DC technology are simplified design, improved reliability and lowered investments costs. Moreover the use of DC in data centres enables a more efficient integration of renewable energy technologies into the power supply design.

(3)

3 Populärvetenskaplig sammanfattning

Datacenter runt om i världen står för ett energianvändande i samma storleksordning som flyget. Fokus har tidigare lagts främst på säkerhet vilket ibland har medfört ineffektiv kraftförsörjning av serverhallarna. Men stigande elpriser samt krav på minskade

koldioxidutsläpp har lett till att alternativa lösningar för kraftförsörjningen har kommit upp för diskussion. En alternativ lösning är att använda sig av likström istället för växelström i datahallen och på så vis reducera antalet konverteringar mellan växelström och likström.

I ett typiskt datacenter återfinns ett antal servrar som strömförsörjs via en UPS

(uninterruptible power supply). I UPSen omvandlas inkommande 230 V enfas växelspänning till likspänning och passerar sedan via ett batteri eller ett svänghjul för att sedan omvandlas tillbaka till 230 V växelspänning. Därefter matas servrarnas nätaggregat (PSU, power supply units) med växelspänning som i sin tur omvandlar den till 5 och 12 V likspänning. Tanken med den alternativa likströmslösningen är att endast likrikta växelströmmen i ett steg, låta den passera ett batteri för att sedan förse servrarnas nätaggregat med likström. Syftet med detta examensarbete är att bestämma om besparingar är möjliga och hur stora dessa i sådana fall är.

Bakgrunden till projektet är att Energimyndigheten i Eskilstuna installerat ett datacenter med utrustning från IBM som enbart strömförsörjs av en DC UPS levererad av förtaget Netpower Labs. I samband med detta ville man utreda vilka besparingar som var möjliga med denna nya teknik. Examensarbetet skrevs i samarbete med IBM Svenska och Netpower Labs.

Det finns idag flertalet datacenter som drivs på likström. Det finns dock inga nätaggregat som är konstruerade för DC-drift. Istället matar man AC nätaggregat med likström och de fungerar problemfritt. IBM tillhandahåller ett datarack för lagring kallat XiV Storage System, här kallat XiV, som har används för den praktiska delen av detta examensarbete. Ett likadant stativ står i Energimyndighetens datacenter i Eskilstuna. Genom mätningar på olika UPS och PSU för växelström respektive likström togs verkningsgrad vid olika laster för apparaterna fram.

Därefter beräknades elanvändningen under ett år för tre olika kraftförsörjningsalternativ av en XiV, 1) traditionell design(AC UPS och AC PSU matad med AC), 2) Nuvarande DC design (DC UPS och AC PSU matad med DC) samt 3) optimerad DC design (DC UPS och DC PSU).

Ett försök att bygga om en AC PSU till DC/DC drift gjordes eftersom det inte finns några DC PSU tillgängliga för XiV Storage System. Detta försök misslyckades och istället ledde

intervjuer med personer med mångårig erfarenhet av elektronikdesign till en uppskattad verkningsgrad av en optimerad DC PSU.

Alla datacenter kräver kylning i någon form och därför togs underlag för beräkning av kylåtgången fram. Därefter beräknades koldioxidutsläpp och kostnader för drift av XiV- racket. Slutsatsen blev att det går att spara 25 % med alternativ 3 jämfört med alternativ 1 och 11 % med alternativ 2 jämfört med alternativ 1.

Efter genomförda mätningar och beräkningar sattes frågan i ett större perspektiv. På

energimyndigheten i Eskilstuna har man även valt att installera solceller som försörjer delar av datahallen under sommarhalvåret. Kraftförsörjningssystemet för datacenter blir mer effektivt om likström används istället för växelström vid integrering av förnyelsebara

(4)

energikällor såsom solceller eller vindkraft i ett eftersom färre omvandlingar mellan likström och växelström behöver göras.

En annan aspekt av att använda likström istället för växelström i datacenter är en förenklad och mer tillförlitlig strömförsörjning. Likström har i jämförelse med växelström färre kontrolparametrar vilket medför färre komponenter för likströmslösningen och därmed färre delar som kan gå sönder. Färre komponenter leder också till lägre investeringskostnader samt att apparaturen tar mindre plats.

(5)

5

1. Introduction

Data centres around the world use about as much energy as the air traffic¹. Energy related costs of a data centre stand for about 12 % of the data centre expenditure², the second largest expenditure after personnel costs. The increasing need for larger storage and server capacities combined with rising electricity prices and requirements on lowered carbon dioxide emissions result in a need to lower the power usage in the data centres.

If the total losses in the equipment in the data centres could be lowered the power usage would be reduced. A proposed solution to achieve a more efficient power supply of the data centres is to use direct current (DC) converted directly from the grid instead of converting the current in several steps between AC (alternating current) and DC.

IBM provides a server rack for storage called XiV Storage System, here referred to as XiV.

The Swedish Energy Agency in Eskilstuna has installed one XiV from IBM to be powered by a DC solution supplied by a Swedish company named Netpower Labs AB.

1.1 Objective of Thesis

The thesis is written in co-operation with IBM Sweden and Netpower Labs. The objective was to measure and compare the efficiency of a traditional AC solution and a DC UPS solution powering the XiV. Thereafter the possible reduction of energy usage when converting to DC in an XiV server rack compared to the same rack powered with AC was calculated.

The issue was then considered in a wider perspective. What benefits or complications come with the DC technology? How can this technology be integrated with renewable energy sources? What other studies have been carried out in the same field of study?

1.2 Method

The most justifying method for determination of the possible reduction in energy use in a server using DC would be to measure two identical servers with identical load, one fed with AC and the other with DC, and measure the energy usage during a year. But due to financial and time limitations the energy usage was determined through measurements of the efficiency of the power supply units (PSU) and uninterruptible power suppliers (UPS). The

measurements were made in Netpower Labs Laboratory in Nacka, Stockholm and Rimasters Laboratory in Söderhamn north of Stockholm.

Thereafter interviews and literature studies gave a base for calculations and other aspects of using direct current instead of alternating current in data centres.

1.3 How to Read this Report

In the main part of the report results, supposed savings in energy and carbon dioxide emissions and some other considerations related to DC usage in data centres are presented.

The measurement arrangements, detailed results, calculations and some theoretical background on alternating and direct current are presented in the appendices.

(7)

7

2. Background

2.1 Traditional Data Centre

Figure 1 shows how a traditional data centre is powered. The standard input in Sweden is 3- phase 400 V AC from a transformer connected to the grid and single phase 230 V AC to appliances i.e. servers in a data centre. Big data centres are often connected directly to dedicated 12 kV high voltage lines.

Figure 1. Schematic sketch of the power supply system in a traditional data centre with uninterruptible power supply and power supply units powering the servers.

A centralized UPS powers all the data equipment with single phase AC. The UPS can be either a battery or a kinetic flywheel. There is also a cooling system maintaining the

temperature around 25 °C. Most data centres also have diesel generators for back-up power.

2.1.1 The XiV Storage System

IBM’s XiV Storage System contains between 6 and 15 servers and 3 uninterruptible UPS powering the servers. The three UPS’s share the load to the servers between them. There are are two PSU in each server. The two power supply units share the load between them.

In order to compare the AC power supply with a DC solution the measurements have been limited to the storage rack PSU and UPS.

AC/DC DC/DC ^Load

PSU

UPS Grid

connection 230

V AC Battery

12 V, 5V Reserve

power 400 V/230 V AC

Server

AC/DC DC/AC

(8)

Figure 2. XiV rack with one server and one UPS removed for measurements 2.2 Powering Data Centres with DC

Figure 3 shows the proposed solution of a DC powered data centre. A data centre normally feed from a 3-phase line has an isolating transformer either connected to high voltage (10 kV) or low voltage (0,4 kV) feeding 230 V AC to rectifier modules. Thereafter the current is passed through a battery and after that directly fed to the servers. In the IBM XIV-case with a single phase DC UPS installed in the rack an isolating transformer is not needed.

Figure 3. Schematic sketch of a data centre powered with DC. The conversions marked with red crosses are removed with this alternative design.

2.2.1 The DC Power Supply Unit

Figure 4 shows a schematic sketch of an AC PSU and which steps that can be excluded in a power supply unit built for DC/DC operation. The idea is to remove the bridge rectifier and the filters and preregulators and thereby improve the efficiency of the PSU.

AC/DC DC/DC ^Load

PSU

UPS Grid

connection 350

V DC Battery

12 V, 5V Reserve

power 230 V AC

Server

AC/DC DC/AC

(9)

9

Figure 4. Schematic sketch of a power supply unit fed with DC 2.3 DC Data Centres in Sweden

Netpower Labs has several reference installations where data centres are powered with the company’s DC UPS system. During the writing of this thesis two of these data centres have been visited. They are Compare Computerroom and the Swedish Energy Agency data centre.

The power supply units in the servers are constructed for 230 V AC input power but work as well on 350 V DC³. Since the standard input is 230 V AC to the PSU all installations so far have been standard AC PSU’s fed with DC.

2.3.1 Compare Computerroom, Hammarö, Sweden

Compare Computerroom is located at Sätterstrand Business Park in Hammarö municipality in the western part of Sweden. The Computerroom is a co-location where different companies can rent room for their servers. The Computerroom has been developed as part of a project to build a data centre that is as energy efficient as possible using for example heat recovery and a modern cooling system. The project started with a test laboratory where the DC equipment was tested. The DC UPS has now been integrated in one part of the Computerroom.

2.3.2 Swedish Energy Agency, Eskilstuna, Sweden

The Swedish Energy Agency is located in Eskilstuna south of Stockholm. The DC UPS system was installed in 2010. All the servers and storage servers in the data centre are powered with DC.

3. Measurements

In order to calculate the supposed reduction in electricity usage and thereby cost and carbon dioxide emission that comes with the running of an XiV Storage System on DC the efficiency of the equipment at different loads was measured.

The energy efficiency was determined for the following devices:

1. AC UPS XiV Storage System 2. DC UPS (Netpower Labs)

3. AC PSU fed with AC (XiV PSU) 4. AC PSU fed with DC (XiV PSU)

3 See Appendix 3 Why AC-equipment can be powered with DC

(10)

The results from the experiments are shown in this chapter. For a more detailed description of the measurements please see attached appendices.

There is no DC/DC PSU available for the XiV. An attempt to construct a DC PSU from an AC PSU was made but this experiment failed. Instead interviews with a number of persons with experience of power electronics design were made in order to obtain an estimated efficiency of a DC/DC power supply unit.

3.1 Efficiency of UPS

Figure 5. Efficiency UPS

Figure 5 shows the efficiency of the AC and DC UPS. The measurements of the DC UPS were carried out on one rectifier, which has a maximum power output of 2.3 kW.

60%

65%

70%

75%

80%

85%

90%

95%

100%

0 500 1000 1500 2000 2500 3000 Efficiency

Load [W]

Efficiency UPS

AC UPS DC UPS

(11)

11 3.2 Efficiency of PSU

Figure 6. Efficiency PSU

Figure 6 shows the efficiency of the three different power supply units. The red line shows the AC PSU fed with single phase 230 V AC and the blue line shows the AC PSU fed with 350 V DC. The supposed efficiency for a DC PSU is between 0.9 and 0.95⁴. The green line in the diagram above shows an estimated efficiency for the optimized DC PSU.

3.3 Measurement Inaccuracy

The calculations of the measurement inaccuracies are presented in Appendix 1.6 Measurement Inaccuracy. The inaccuracy of the efficiencies, η, at normal operation are presented below.

Table 1. Intervals for measurement inaccuracy at normal operation (Load: 200 W per PSU, normal operation)

Lowest η Average efficiency

(η) Highest η

AC UPS 0.909 0.911 0.914

DC UPS 0.964 0.966 0.968

AC PSU fed with AC 0.734 0.743 0.751

AC PSU fed with DC 0.771 0.780 0.789

The table shows the interval of uncertainty due to inaccuracy of the measurements equipment used for the determination of the efficiency. As can be seen none of the intervals overlap with the other for the UPS or the PSU respectively. Thereby it can be concluded that the efficiency

4 Louis Masreliez, Powerbox AB, Thomas Sahlström Efore AB, Carl Strandberg Pelikama AB 30%

40%

50%

60%

70%

80%

90%

100%

0 200 400 600 800 1000

Ef+iciency

Load [W]

Ef+iciency PSU

AC PSU fed with AC

AC PSU fed wtih DC

Optimized DC PSU

(12)

for the DC UPS is higher than the efficiency of the AC UPS and that the efficiency for the AC PSU fed with DC is higher than the AC PSU fed with AC.

Furthermore, the instrument specified accuracies might include uncertainties in the calibration and other systematic errors. Since the same instruments were used the accuracy in the

difference of the efficiency may be better than the numbers given.

4. Calculations

All calculations are fully presented in Appendix 2 Calculations.

The servers have a working load of around 500-550 W input under normal conditions.⁵ This implies 250 W input per power supply unit and corresponds to an output to the load of around 200 W per PSU, which is what has been used for the calculations in order to get comparable results of losses in the different power supply solutions during normal operation.

The results from the measurements were used to calculate the power and energy consumption for the different scenarios below.

1. Traditional configuration; AC UPS and AC PSU fed with AC (XiV Storage System) 2. Current DC configuration: DC UPS and AC PSU fed with DC

3. Optimized DC configuration: DC UPS and DC PSU

5. Results

5.1 Power and Energy Usage

Table 2. Results from measurements and calculations for the three configurations

Alternative 1. AC fed with AC

2. AC fed with DC

3.

Optimized DC

Efficiency UPS 0.91 0.97 0.97

Efficiency PSU 0.74 0.78 0.93

Total efficiency 0.67 0.76 0.9

Total Power Usage [W] 5300 4800 4000

Electrical Energy Usage per

year [kWh] 47000 42000 35000

(13)

13 5.2 Energy Usage Including Cooling

At IBM Sweden the common rule of thumb for estimating the cooling needed in a data centre is that one kWh consumed in the servers requires one kWh of cooling. The cooling of a data centre is usually via cold air fed from beneath the floor. The cold air can be produced in many different ways. The most environmentally friendly way is to use free cooling, something that is becoming more and more common in Sweden and other countries with colder climates. The outside temperature is then used to cool the air in the data centre via heat exchangers during wintertime⁶.

The other alternative is to use reverse cycled heat pumps for production of cold water, either delivered from a co-generated power plant or produced by heat pumps in the data centre. This is standard in most of today’s data centres. The coefficient of performance (COP) for a reversed cycle heat pump is the ratio of cooling produced and the electrical input to the compressor. The COP for a standard cooling machine can be estimated⁷ to 4 and thus the additional electricity usage for cooling is 25 % of the power to the servers.

Since the calculation of the cooling needed is used for comparing the different alternatives an estimated 25 % in electricity usage for a reversed cycle heat pump has been added to the different alternatives.

Table 3. Energy usage per year including cooling

Alternative

1. AC fed with AC

3.

Optimized DC Electrical Energy Usage per

year including cooling [kWh] 59000 52000 44000

5.3 Carbon Dioxide Emission

The rule of thumb for calculations of carbon dioxide emission per produced kWh in Sweden is 100 g for Nordic mixture or 1 kg for imported electricity (European mixture)⁸.

Table 4. Carbon dioxide emissions per year

3.

Optimized DC Carbon dioxide emission per

year Nordic mixture [tons]

(including cooling)

5.9 5.2 4.4

Carbon dioxide emission per year imported electricity [tons]

(including cooling)

590 520 440

6 Yvonne Lithner and Jan Christer Sköld, IBM Sweden

7 Johan Bergman, Johnson Controls

8 http://www.energiradgivningen.se/index.php?option=com_content&task=view&id=52&Itemid=30

(14)

5.4 Economy

The price per kWh varies over the year depending on various factors. According to Statistics Sweden (SCB) the average price in 2010 was 0.89 SEK (0.12 USD⁹) per kWh when the costumer consumed 60 MWh annually which corresponds to the annual energy usage of an XiV.¹⁰

Table 5. Costs per year

3.

Optimized DC Costs per year including

cooling [SEK]

52000 (7000 USD)

46000 (6300 USD)

39000 (5200 USD) 5.4.1 Investment Costs

The DC solution will imply fewer components and thereby reduced production costs and possible lowered investment costs. According to a report released by Intel Labs in 2010 the usage of DC in data centres will result in 15 % savings in electrical facility cost.¹¹ The

reduction in cooling needed may result in lowered investment costs for the cooling apparatus.

5.5 Savings

The table below presents the savings in energy usage, energy used including cooling, carbon dioxide emissions and costs during a year of operation with alternative 2 and 3 compared to alternative 1 for one XiV Storage System.

Table 6. Savings in energy usage, carbon dioxide emissions and costs Savings per year 2. AC fed

with DC 3. Optimized DC Electrical Energy Used

excluding cooling [kWh]

5200 (11%) 12000 (25 %) Electrical Energy Used

including cooling [kWh]

6400 15000

Carbon Dioxide Emissions Nordic mixture [tons]

0.7 1.5

Carbon Dioxide Emissions Imported Electricity [tons]

700 1500

Financial Savings [SEK] 5600 (770USD)

13000 (1800 USD)

The total savings are 11 % for alternative 2 compared to alternative 1 and 25 % for alternative 3 compared to alternative 1 as shown in the table above.

9 The Swedish central bank, Exchange rates, mean value USD 2010.

10

(15)

15

6. Other Aspects of Using DC in Data Centres

6.1 Renewable Energy

The Swedish Energy Agency in Eskilstuna has installed solar panels with a maximal output of 12 kW powering the data centre during summertime. There are several benefits using DC in the data centre when integrating renewable energy sources to the power supply system. This since many conversion steps can be removed (see fig 7) compared to the traditional setup where AC is used in the data centre.¹²

Figure 7. Schematic sketch of a DC powered data centre with renewable energy sources integrated into the design.

6.2 Reliability and Safety

Direct current has fewer control parameters compared to alternating current. Alternating current is related to various parameters such as voltage, phase, frequency and waveform whereas direct current is ‘simpler’ with only voltage as control parameter. This leads to simplified design and thereby safer and more reliable operation.¹³ The probability of failure over five years is 6.72 % for the DC solutions compared to 13.63 % for the traditional AC solution according to reliability predictions made by Intel Labs in 2010.¹⁴

The most common argument against the usage of DC is the problem with breaking the current. The first electric grids were all providing 220 or 110 V DC to the households. DC was then replaced with AC because of troubles breaking the DC current, something that sometimes caused fire accidents. But the breaking of the current is only problematic when

12 Savage, Nordhaus, Jamieson, ”DC grids: Benefits and Barriers”, YALE School of Forestry and Environmental Studies, 2010, page 54,

13 Murrill, Sonnenberg, “Evaluating the Opportunity for DC Power in the Data Center”, Emerson Network Power

14Albridge, Pratt, Kumar, Dupy, AiLee, “Evaluating 400 V Direct-Current for Data Centers”, Intel Labs 2010, page 11

Load

AC/DC DC/DC

PSU

UPS Grid

connection

Battery Reserve

power

Server

AC/DC DC/AC

DC/DC DC/AC

(16)

inductive loads (toasters, heat radiators) are powered with DC. Most electric equipment in today’s data centres are capacitive or resistive loads (computers, compact fluorescents lamps etc.), also called electronic loads. Since the inductance in modern electronic loads is much lower compared to old inductive loads there is no discharge when breaking the current.¹⁵

7. Other Studies in the Same Field

Lawrence Berkley National Laboratory published a report together with Sun Microsystems named ‘DC Power for Improved Data Center Efficiency’ in March 2008. They measured different setups of AC and DC powered data centres during a year of operation and concluded that an energy efficiency improvement of up to 28 % can be achieved when converting to a DC solution.

Both Emerson Network Power and Intel have also published several reports and studies in the field of DC in data centres and DC micro grids and concluded that energy efficiency

improvements are possible when converting to DC operation.

8. Discussion

The preferred way of determining the gains in efficiency improvement would be to measure energy usage of an AC and a DC powered server with identical load during a year. This is how the measurements at Lawrence Berkley National Laboratory were carried out. The result from this study was an energy efficiency improvement of 28 % compared to 25 % in this study. Therefore the results from the measurements made in this Master Thesis are believed to be accurate.

The reason for the higher efficiency for an AC PSU fed with DC compared to the same apparatus powered with AC is that higher voltage entering the rectifier bridge in the PSU implies lowered losses. Higher input voltage implies lower current and thereby lowered losses in the diodes in the PSU.

The need for more efficient data centres in order to reduce costs and carbon dioxide emissions will be in focus in the near future. Within the data centre business the focus has been to achieve a more efficient and environmentally friendly cooling of the data centre. But this new approach and the high reduction in energy use and the simplicity of the design will probably result in DC powered data centres being more and more frequent.

Prices for electricity and carbon dioxide emissions per kWh vary a lot between different countries. In Sweden hydro and nuclear power plants stand for the majority of the electricity production. For data centres in other parts of the world where fossil fuels stand for the

majority of the electrical energy production the carbon dioxide emissions per kWh are higher and similar to the 1 kg per kWh used for the carbon dioxide emissions calculations.

The savings presented in the report are calculated for one XiV Storage rack. In larger data centres the power consumption can be several MW. For example IBM’s data centre in Kista has a power consumption of 1.6 MW. The savings in costs and carbon dioxide emissions for running the facility will thereby be about 300 times the savings for one XiV.

(17)

17

9. Conclusion

This Master Thesis and other investigations and measurements indicate that up to 25-28 % energy efficiency improvement is possible with the DC-technology compared to traditional design. This will also result in 25 % reduction of operational costs and carbon dioxide emissions from the data centre.

The DC power supply solution is also easier and more efficient when integrating renewable energy sources into the data centre power supply system. Other aspects such as higher

reliability and fever components will result in less component failure and lowered investment costs.

Since data centres around the world use about as much energy as the air traffic the results from this report indicate that the environmental impacts of the data centres can be lowered significantly on a global basis.

(18)

List of references

Personal contacts

Lija, Magnus, IBM Sweden Lithner, Yvonne, IBM Sweden Masreliez, Louis, Powerbox AB Sahlström, Thomas, Efore AB Starenby, Patrik, IBM Sweden Strandberg, Karl, Pelikama AB Telephone interviews

Bergman, Johan, Johnson Control Sköld, Jan-Christer, IBM Sweden Literature and reports

Albridge Tomm, Pratt Annabelle, Kumar Pavan, Dupy Dwight, AiLee Guy, “Evaluating 400 V Direct-Current for Data Centers”, Intel Labs 2010

Fortenbery Brian, Ton My, Tschudi William ‘DC Power for Improved Data Center Efficiency’ Lawrence Berkley National Laboratories, 2008

Murril Mark, Sonnenberg JB, ‘Evaluating the Opportunity for DC Power in the Data Center’, Emerson Network Power

Savage Paul, Nordhaus Robert R, Jamieson Sean P, ”DC grids: Benefits and Barriers”, YALE School of Forestry and Environmental Studies, 2010.

Articles

”Data centre energy costs to surge”, published on Techworld.com 04 October 10 Internet

”Gartner Says Data Centres Account for 23 Per Cent of Global ICT CO2 Emissions”

http://www.gartner.com/it/page.jsp?id=530912, 2011-06-14

http://www.energiradgivningen.se/index.php?option=com_content&task=view&id=52&Itemi d=30, 2011-05-10

The Swedish central bank

http://www.scb.se/Pages/TableAndChart____212961.aspx (category IB, 20-500 MWh), 2011- 05-10

(19)

19

Appendix 1.1 Efficiency of AC UPS

Background

In the XiV there are three AC UPS. One of these has been used for measurements of the efficiency factor.

Figure A1.1:1. The UPS used for measurements of the efficiency Measurement arrangements

In the experiment an instrument for power measurements named Yokogawa was borrowed from the Swedish Energy Agency. This instrument was used to measure the incoming power and the Voltech Power Analyser was used to measure the power consumed by the load (see figure A1.1:2).

The loads used in the experiment are heat radiators, one 1150 W, one with variable load 200, 400 and 600 W, and six 400 W radiators. In order to get comparable values of the consumed power the all values were logged one minute after the loads were connected.

Figure A1.1:2. Measurement arrangements of AC UPS Results

The results are presented with decimal comma instead of decimal point.

AC UPS

Yokogawa Power Meter 230 V AC

Voltech Power Analyser 230 V AC

Variable load (Heat radiators)

(20)

Table A1.1:1. Efficiency AC UPS

Power in [W] Yokogawa Power out [W] Voltech Efficiency

299 201 0,672

494 395 0,800

693 589 0,850

1113 993 0,892

1523 1380 0,906

1925 1760 0,914

2325 2131 0,917

2666 2447 0,918

3083 2828 0,917

3483 3186 0,915

The load per UPS is 1617 W (see Appendix 2 Calculations) at normal operation. In order to determine the efficiency at this load a linear relationship between the two marked efficiencies in the table above was assumed. The linear relationship is written as follows:

! = 2.158 ∙ 10^!!! + 0.876

where y is the efficiency and x the load. The efficiency at 1617 W is therefore

! = 2.158 ∙ 10^!!∙ 1617 + 0.876 = 0.9112 … ≈ 0.9112 The input to the UPS is with the same method

!_!" = 1617

0.9112= 1774.55 … ≈ 1775

(21)

21

Appendix 1.2 Efficiency of AC PSU fed with AC

Background

In each XiV there are between six and twelve servers and two Power Supply Units (PSU) in each server. The UPS feeds the servers with single phase AC 230 V that is converted to DC 12 V and 5 V.

Figure A1.2:1. Measurement setup for the AC power supply unit Measurement with Yokogawa Power Meter

Figure A1.2:2. Schematic sketch of measurement arrangements

The Yokogawa instrument was used to measure the incoming power and the two multimeters were put on the secondary side.

Results

In the table presenting the results decimal comma is used instead of decimal point.

230 V AC

Yokogawa Power Meter

AC PSU

Current meter Variable load

Voltmeter

12 V

(22)

Table A1.2:1 Efficiency AC PSU fed with AC

Pin [W] Uout [V] Iout [A] DVM92 Load [W] Efficiency η

72,5 12,4 2,1 26, 0,359

93,7 12,41 3,6 45 0,477

134,6 12,41 6,5 81 0,599

157,5 12,41 8,1 101 0,638

217,8 12,42 12,4 154 0,707

274,6 12,43 16,4 204 0,742

316,4 12,44 20 249 0,786

381,5 12,45 24,6 306 0,803

442,6 12,46 29 361 0,816

496 12,47 32,7 408 0,822

556 12,48 37 462 0,831

627 12,49 42 525 0,837

713 12,51 48,1 602 0,844

827 12,52 55,9 700 0,846

944 12,54 64 803 0,850

1033 12,55 70 879 0,850

The efficiency and load marked with yellow above is where the PSU lies in normal operation

(23)

23

Appendix 1.3 Efficiency of AC PSU fed with DC

Background

The servers at the Swedish Energy Agency in Eskilstuna have AC power supply units fed with DC. The motive for measuring the efficiency of an AC PSU fed with DC was to determine whether feeding them with 350 V DC reduces the losses in the PSU.

Measurement arrangements

Figure A1.3:1. Schematic sketch of measurement arrangements The Voltech Power Analyser was used to measure the incoming power and the two multimeters were put on the secondary side.

Results

The results are presented with decimal comma instead of decimal point.

Table A1.3:1 Efficiency AC PSU fed with AC

Pin [W] Iout [A] Uout [V] Load [W] Efficiency η

63,91 2,1 12,34 26 0,406

83,72 3,6 12,39 45 0,533

123,63 6,5 12,4 81 0,652

145 8,1 12,4 100 0,693

204,7 12,4 12,41 154 0,752

261,1 16,4 12,42 204 0,780

310,8 20 12,43 249 0,800

377,1 24,6 12,44 306 0,812

439,9 29 12,45 361 0,821

493,5 32,7 12,46 407 0,826

554 37 12,47 461 0,833

623,9 42 12,48 524 0,840

709 48,1 12,49 601 0,847

825,9 55,9 12,51 700 0,847

944,7 64 12,53 802 0,849

1038 70 12,54 879 0,846

The efficiency and load marked with yellow above is where the PSU lies in normal operation

350 V

DC

Voltech Power Analyser

AC PSU

Currentmeter Variable load

Voltmeter

12 V

(24)

Appendix 1.4 Rebuilt AC Power Supply Unit

This document is kept at Netpower Labs AB Adress:

Stubbsundsv. 17 13141 Nacka Stockholm Sweden

Contact person: John Åkerlund

(25)

25

Appendix 1.5 Efficiency of DC UPS

Background

The DC UPS consist of a number of rectifiers and batteries. The input to the rectifier can vary between 230-270V AC. The output to the batteries is 350 V. Since the current flows through the batteries in normal operation the only losses are in the rectifier, there are small losses in the batteries due to leakage currents but these are considered to be negligible.

An input of 265V-270 V gives the highest efficiency of the rectifier¹⁶. Therefore when possible the transformer should be changed to give 270 V output instead of 230 V which is the standard.

First a series of measurements was carried out in Nacka at Netpowers Laboratory with 230 V input to the rectifier. In order to measure the efficiency with different inputs between 230 and 270 V a second measurement on the rectifier was made in Söderhamn north of Stockholm at Rimasters laboratory where a variable AC input is available.

Measurement in Söderhamn

At the laboratory in Söderhamn the efficiency of the rectifier was tested with inputs of between 228, 230, 240, 250, 260, 265 and 270 volt.

Figure A1.5:1. Measurement equipment in Söderhamn

16 John Åkerlund, Netpower Labs

(26)

Figure A1.5:2. Rectifier The measurement arrangements are shown below.

Figure A5:3. Measurement arrangements Results

Table A1.5:1 and A1.5:2 show the efficiency at 230 and 265 V respectively. The efficiencies at 265 V input to the rectifier have been used for the calculations since this input gives the highest efficiency. Throughout the tables in this appendix decimal comma will be used instead of decimal point.

230 -‐

265 V

Voltech Power Analyser

DC UPS

FLUKE 45 Digital multimeter (currentmeter)

Variable load

Agilent 34401A (voltmeter)

350 V

(27)

27 Table A1.5:1. Efficiency at 230 V input

Table A1.5:2. Efficiency at 265 V input

The load per UPS is around 1200 W in normal operation. Therefore the efficiency marked with yellow in the table above has been used for further calculations.

230 V

Current out

[A] Voltage [V] Load [W]

Power in

[W] Efficiency

0,42 353,09 148 164 0,904

0,849 352,84 300 320 0,938

1,699 352,31 599 628 0,953

2,498 351,83 879 915 0,961

3,408 351,3 1197 1245 0,962

4,306 350,83 1511 1571 0,962

5,185 350,23 1816 1893 0,959

6,008 349,01 2097 2192 0,957

6,626 345,94 2292 2402 0,954

265 V I out [A] U out [V] Load [W]

Power in

[W] Efficiency

0,422 353,18 149 163 0,917

0,856 352,93 302 321 0,940

1,699 352,4 599 623 0,962

2,514 351,98 885 920 0,962

3,408 351,32 1197 1240 0,966

4,301 350,85 1509 1562 0,966

5,207 350,33 1824 1889 0,966

6,008 349,17 2098 2176 0,964

6,605 346,06 2286 2375 0,962

7,328 340,1 2492 2593 0,961

(28)

Appendix 1.6 Measurement Inaccuracy

In this section the measurement inaccuracies for the efficiency at normal operation are presented.

The efficiency is calculated from the following formula η = !_!"#

!_!"

where η is the efficiency, Pout the outgoing power from the UPS and Pin the incoming power to the UPS.

The logarithm of the expression above gives

ln η = ln !_!"#− ln !_!"

differentiation gives

!η

η = !!_!"#

!_!"# − !!_!"

!_!"

The standard deviation uη is then calculated from the formula

!_! η

! = !_!_!"#

!_!"#

!+ !_!_!"

!_!"

!

1.6.1 AC UPS

The accuracy rates for the instruments used are

!_!!"#

!!"# = 0.2 % = 0.002 (Voltech PM100)

!_!!"

!_!" = 0.2 % = 0.002 (Yokogawa WT200 Digital Power Meter)

The standard deviation is thereafter calculated

!!

!

! = 0.002 ^!+ 0.002 ^! = 8 … ∙ 10^!! (∗)

The efficiency η is 0.9112… and the standard deviation is

!_! = ± 0.9112 ∙ 8 ∙ 10^!!= ±0.00257 … ≈ ±0.00258 (∗∗) which results in an uncertainty of the efficiency in the interval of

(0.909; 0.914) 1.6.2 AC PSU fed with AC

!_!!"#

= 0.8 % = 0.008 (BS3604W)

(29)

29

!_!!"#

!_!"# = 0.8 % ¹⁷= 0.008 (Velleman DVM92)

!_!!"

!_!" = 0.2 % = 0.002 (Yokogawa WT200 Digital Power Meter)

The standard deviation is thereafter calculated from the accuracy rates given above and the formulas (*) and (**) used in 1.6.1. The efficiency, η, is 0.742 at 200 W load per PSU (normal operation). This results in an uncertainty of this efficiency in the interval of

(0.734; 0.751) 1.6.3 AC PSU fed with DC

!_!!"#

!_!"# = 0.8 % = 0.008 (BS3604W)

!_!!"#

!_!"# = 0.8 % = 0.008 ¹⁸ (Velleman DVM92)

!_!!"

!_!" = 0.2 % = 0.002 (Voltech Power Analyser PM100)

The standard deviation is thereafter calculated from the accuracy rates given above and the formulas (*) and (**) used in 1.6.1. The efficiency is η is 0.780 at 200 W load per PSU (normal operation). This results in an uncertainty of this efficiency in the interval of

(0.771; 0.789) 1.6.4 DC UPS

!_!!"#

!!"# = 0.0035 % = 0.000035 (Aglient 34401A)

!_!!"#

!_!"# = 0.05 % = 0.0005 (FLUKE 45 Digital Display)

!_!!"

!_!" =0.2 % = 0.002 (Voltech PM100)

The standard deviation is thereafter calculated from the accuracy rates given above and the formulas (*) and (**) used in 1.6.1. The efficiency is η is 0.966 at normal operation. This results in an uncertainty of this efficiency in the interval of

(0.964; 0.968)

17 0.8 % accuracy is at normal operation; the instrument resolution has been neglected.

18 0.8 % accuracy is at normal operation; the instrument resolution has been neglected.

(30)

Appendix 2 Calculations

In each XiV there are between 9 and 15 servers and therefore between 18 and 30 power supply units. The Energy Agency in Eskilstuna currently uses 9 servers in their XiV and therefore the energy usage has been calculated for this configuration.

At normal operation the input per server is approximately 500-550 W¹⁹ which corresponds to a load of 200 W per PSU.

The results from the measurements were used to calculate the energy used for the different scenarios shown below.

The results are summarized in Table 2 under 5.1 Power and Energy Consumption.

DC UPS

A proposed solution of a DC power supply in the XiV is to use 4 rectifiers connected in parallel pairs to the batteries. The maximal load on one rectifier is 2500 W.

Calculations

XiV Storage System

Current DC configuration

Optimal DC configuration

Load per PSU [W] 200 200 200

Efficiency PSU 0.74 0.78 0.93

Total input per server [W] 539 513 430

No. of servers 9 9 9

Total input to all servers [W] 4851 4615 3871

Load to UPS [W] 4851 4615 3871

No of UPS 3 4 4

Load per UPS [W] 1617 1154 968

Efficiency UPS 0.91 0.97 0.97

Input to rack [W] 5331 4758 3991

Annual Electrical Energy Usage

Power Usage [W] 5331 4758 3991

No of hours per year 8760 8760 8760

Energy Usage

per year [kWh] 46699 41681 34958

(31)

31

Appendix 3. Why AC-equipment can be powered with DC

1. Electronic loads

All AC devices classified as electronic loads have a bridge rectifier where the incoming alternating current is converted to DC. An electronic load is for example a computer, compact fluorescent lamp, induction cooker etc. The electronic loads need lower voltages for example 5 and 12 V DC to be converted from the incoming grid AC single phase.²⁰

Figure A3:1. Bridge rectifier showing alternating current directions²¹ In Sweden the standard voltage of single phase AC is 230 V. When passing the bridge

rectifier the current is converted to 320 V DC.²² When feeding the bridge rectifier with 350 V DC the voltage passes over two opposite diodes depending on how the poles are connected to the input side.²³ All electronic devices have an input voltage interval due to fluctuations in the grid and therefore the output from the rectifier can vary between around 260 V and 380 V.

Consequently any device classified as electronic load can be powered directly with DC. All of the equipment in a data centre such as servers and fluorescent lighting can run on DC without any complications.

20 Karl Strandberg, Pelikama AB

21 Picture source: Sears, Francis W., Mark W. Zemansky and Hugh D. Young, University Physics, Sixth Ed., Addison-Wesely Publishing Co., Inc., 1982, p. 685.)

Energy Efficiency Improvements Using DC in Data Centres