ENERGY PERFORMANCE OF DATA CENTRES IN SINGAPORE SUN HANSONG. (B.Eng.) A THESIS SUBMITTED FOR THE DEGREE OF MASTER OF SCIENCE (BUILDING)

Transcription

1 ENERGY PERFORMANCE OF DATA CENTRES IN SINGAPORE SUN HANSONG (B.Eng.) A THESIS SUBMITTED FOR THE DEGREE OF MASTER OF SCIENCE (BUILDING) DEPARTMENT OF BUILDING SCHOOL OF DESIGN AND ENVIRONMENT NATIONAL UNIVERSITY OF SINGAPORE 2004

2 ACKNOWLEDGEMENT I would like to express my deepest gratitude to Associate Professor Lee Siew Eang for his support, guidance and valuable advice throughout this academic exercise. I would like to thank Dr. Schafer for his support, valuable advice and expertise during the measurements. I would like to express my appreciation to Mr. Majid for the assistance, advice and friendship. I would like to thank Dr. David Ho for his guidance and advice on neuro-fuzzy network modeling. My appreciation and thanks to the following people who have been helpful and for their great contributions in the accomplishment of the study: Data centre managers and personnel for giving accesses to the facilities and their assistances during the field measurements. The members of Energy Sustainability Research Unit (ESRU) of the Centre for Total Building Performance (CTBP), for sharing their invaluable knowledge and experiences during the group discussion. Finally, thanks to my friends whom I have been working with throughout the period of study in the National University of Singapore. i

3 TABLE OF CONTENTS ACKNOWLEDGEMENT i TABLE OF CONTENT....ii SUMMARY...vi List of Tables.....ix List of Figures...xi CHAPTER 1 INTRODUCTION Background Objectives of the study Scope and limits Outline of the thesis..6 CHAPTER 2 REVIEW OF LITERATURE ON DATA CENTRE ENERGY PERFORMANCE Introduction Data centre and its classification Data centre common designs Typical energy consuming systems Infrastructure and space use Occupants Environmental conditions Power quality and reliability Data centre energy performance Data centres put substantial burden on utilities Environmental concerns High designed power demands of data centres Actual energy consumption of data centres Breakdown of data centre energy consumption Over-designing and inefficient energy use of data centres Reasons for data centre over-designing Difficulties of forecasting data centre power density Indicators of data centre energy performance Total data centre power density (W/m 2 ) and energy consumption index (kwh/m 2 /year) Energy consumption of IT equipment as a percentage of total data centre energy consumption Relative energy effectiveness of HVAC system Conclusion...35 ii

4 CHAPTER 3 METHODOLOGY AND RESEARCH DESIGN Introduction Preliminary modeling of data centre energy performance Model of data centre energy consumption Model of IT equipment energy performance Model of HVAC system energy performance Model of UPS energy performance Model of lighting energy performance Metrics of data centre energy performance Methodology of field measurements Conversation with data centre operators Expert walk-through and preliminary assessment Environmental measurement Energy consumption measurement Instrumentation Calculation of energy consumption and power demand Selection of data centres Uncertainty analysis of data measured Verification of data centre energy performance model Benchmarking of data centre energy performance Neuro-fuzzy network modeling of data centre energy performance Conclusions CHAPTER 4 MEASUREMENTS AND RESULTS Introduction General description of data centres studied Results of electricity energy consumption measurements Profile of power demands of data centre Profile of power demands of data centre Profile of power demands of data centre Profile of power demands of data centre Profile of power demands of data centre Profile of power demands of data centre Uncertainty analysis of data obtained Estimation of annual energy consumption Power density based on gross floor area of data centre Energy consumption per unit gross floor area of data centre Discussion Overall data centre energy performance Total data centre power demand density Total data centre energy consumption Over-designing of total power demand of data centre Breakdown of energy consumption of data centre Percentage of energy consumption of IT equipment Percentage of energy consumption of HVAC system Percentage of energy consumption of UPS losses IT equipment energy performance IT equipment power demand density..95 iii

5 IT equipment power density based on footprint area Actual power demand vs. nameplate rated power demand of IT equipment Over-designed power demand of IT equipment UPS energy performance Efficiency of UPS HVAC system energy performance Indoor environmental conditions Influences of outdoor weather conditions HVAC system power demand and energy consumption Relative energy effectiveness of HVAC system Over-designing of HVAC system Evaluation of HVAC system energy performance Lighting system energy performance Lighting system power demand Energy performance other auxiliary systems Wider benchmarking of data centre energy performance with comparative studies Total data centre power density IT equipment power density HVAC system power density IT equipment power demand as percentage of total data centre power demand Relative energy effectiveness of HVAC system Over-designing of IT equipment power demand General recommendations of improving data centre energy performance Conclusions CHAPTER 5 PARAMETER ANALYSIS OF DATA CENTRE ENERGY PERFORMANCE Introduction Parameter analysis of IT equipment energy performance Parameter analysis of UPS energy performance Parameter analysis of HVAC system energy performance Parameter analysis of lighting system energy performance Parameter analysis of total data centre energy performance Conclusions CHAPTER 6 NEURO-FUZZY NETWORK MODELING OF DATA CENTRE ENERGY PERFORMANCE Introduction Application of neuro-fuzzy network in modeling data centre energy performance Knowledge background of neuro-fuzzy network Structure of fuzzy logic model Principles of neural network Neuro-fuzzy network iv

6 6.3.4 Method of neuro-fuzzy network modeling using FuzzyTech Fuzzification Fuzzy inference Defuzzification Neuro-fuzzy training Summary of setting of neuro-fuzzy network modeling Objectives of neuro-fuzzy network modeling of data centre energy performance Modeling methodology Parameter analysis & establishment of fuzzy logic model Neural network training Model verification Evaluation of energy performance and prediction of energy saving potential using neuro-fuzzy network model Conclusions CHAPTER 7 CONCLUSIONS Review and achievement of research objectives Contributions of the study Recommendations of future studies..185 BIBLIOGRAPHY..186 APPENDICES Appendix A: Common energy consuming components of data centre Appendix B: Questionnaire and forms 192 Appendix C: Introduction of TEM Appendix D: Samples of measured data centre power demands.201 Appendix E: Specifications of HVAC and UPS..207 Appendix F: Correlation between outdoor air temperature and HVAC system power demand.211 Appendix G: Specifications of electrical power supply circuits of HVAC and UPS systems.213 Appendix H: Monitored indoor environmental conditions.215 Appendix I: Configuration and terms definitions of neuro-fuzzy training 217 Appendix J: Training data of neuro-fuzzy network model and defined ranges of input and output variables.220 v

7 SUMMARY Data centres, the centralized facilities housing large number of IT equipment to perform various functions of storage, management, processing, and exchange of digital data and information, are essential support for Information and Communication Technology (ICT), which involves universal adoption of computing equipment and network as tools to provide information-based services in the modern knowledge economy. This thesis describes the study on the energy performance of data centres in Singapore. The primary objectives of this study are to determine an empirical energy usage pattern of data centres under tropical climate and to develop whole-system based models of data centre energy performance. Case studies and field measurements are conducted in the six data centres in Singapore. The energy performances of various energy consuming systems of data centres are examined. These include IT equipment, uninterruptible power supply (UPS), HVAC system, lighting and other auxiliary systems. The design and operation strategies of data centres, as well as the energy saving potentials are explored. This study will benefit data centre operators by providing the detailed and accurate energy consumption and power demand information, as well as the recommendations for improving data centre energy performance. The benchmarking of energy performance carried out among the six data centres studied provides the data centre managers or owners with a better understanding of the position of his/her facilities. In vi

8 this way, they may be inspired or motivated to better the facility s efficiency in the process. The results of the study will also aid designers of future data centres to make informed decisions about design and construction of data centres. The methods of accurate energy consumption measurement, evaluation and modeling of data centre energy performance using neuro-fuzzy network approach introduced in this study will be valuable to researchers in the related field. The main results of this study are outlined as follows. Firstly, this study reveals that data centres are highly intense energy consuming facilities. Power demands and energy consumptions per unit gross floor area of data centres in terms of W/m 2 and kwh/m 2 /year are significantly higher than those of normal commercial office spaces in Singapore. Secondly, over-designed power demands, over-sizing of HVAC and UPS systems, low occupancy rate of the gross floor area, poor layout of IT equipment and supporting systems, and unreasonable environmental setting conditions are universal problems of data centre facilities. Significant energy saving potential can be expected. Thirdly, there are significant variations among the six data centres investigated, with respect to the energy consumptions, power demands, operational energy efficiencies of systems, as well as the design and operation strategies. These reflect the lack of design standard and guidelines for data centres. Lastly, on the basis of the case studies and field measurements, dynamic whole-system vii

9 based models of data centre energy performance are developed using neuro-fuzzy network. The models incorporate various energy consuming systems and variables affecting energy performance of data centre. The quantitative prediction of energy saving potential of data centres can be performed. viii

10 List of Tables Table 2.1 Measured data centre power demands from various sources...21 Table 2.2 Reviewed breakdown of end-use components of data centre energy consumptions Table 3.1 Metrics of data centre energy performance.. 46 Table 3.2 Specifications of measuring instruments.. 52 Table 3.3 Example of energy data recorded by data loggers of TEM-1 54 Table 3.4 Backgrounds of selected data centres Table 4.1 Breakdown of data centre gloss floor area Table 4.2 Areas of perforated tiles...64 Table 4.3 Configurations of HVAC systems in data centres 68 Table 4.4 Measured power demands (kw) of data centre Table 4.5 Measured power demands (kw) of data centre Table 4.6 Measured power demands (kw) of data centre 3.71 Table 4.7 Measured power demands (kw) of data centre Table 4.8 Measured power demands (kw) of data centre Table 4.9 Measured power demands (kw) of data centre Table 4.10 Potential errors from measuring instruments (%) Table 4.11 Systematic uncertainties of power demands (kw) of the major energy consuming systems...80 Table 4.12 Mean power demands (kw) of data centres with combined errors Table 4.13 Estimated energy consumptions of data centres (MWh/year) Table 4.14 Power demand densities based on the gross floor area (W/m 2 )..83 Table 4.15 Energy consumptions per unit of gross floor areas (kwh/m 2 /year) 84 Table 4.16 Total designed power demand densities of data centres...88 ix

11 Table 4.17 Detailed breakdown of end-use components of data centre energy consumption. 90 Table 4.18 Actual power demands vs. nameplate rated power demands of IT equipment..97 Table 4.19 Actual power demand vs. designed power demands of IT equipment...98 Table 4.20 Measures of UPS energy efficiency.100 Table 4.21 Measured indoor environmental conditions of data centres studied..104 Table 4.22 Calculation of thermal load balance in data centre Table 4.23 Over-designing of HVAC system 110 Table 4.24 Evaluation of HVAC systems energy performances. 111 Table 4.25 Lighting performance of data centres Table 4.26 Comparison of lighting system energy performance of data centres Table 4.27 Observed opportunities of improving energy efficiencies in data centres Table 5.1 IT equipment footprint areas in data centres..127 Table 6.1 Comparison between actual and predicted energy performances of data centres studied 172 Table 6.2 Estimation of energy saving potential by modifying over-designing of IT equipment power demands of data centres.178 Table 7.1 Summary of overall energy consumptions and power demands of data centres.181 x

12 List of Figures Figure 2.1 Energy consumption distribution in a typical data centre...14 Figure 2.2 Measured UPS efficiency from case studies...24 Figure 2.3 Heat density trends of computing hardware Figure 3.1 Systematic flow chat of research design.37 Figure 3.2 Constitution of data centre energy consumption Figure 3.3 Model of IT equipment energy performance Figure 3.4 Model of HVAC system energy performance...41 Figure 3.5 Model of UPS system energy performance Figure 3.6 Model of lighting system energy performance Figure 3.7 Procedure of field measurement of data centres..48 Figure 3.8 Setting up of energy consumption measurement Figure 3.9 Electronic thermal meter Figure 3.10 Sling thermal meter Figure 3.11 Lighting level meter Figure 3.12 HOBO meter Figure 3.13 True RMS current meter...53 Figure 3.14 True energy meter TEM Figure 4.1 Perforated and grilled raised floor Figure 4.2 IT equipment in data centres...65 Figure 4.3 Poor cable management in data centres..65 Figure 4.4 HVAC systems of data centres 68 Figure 4.5 Monitored total power demand of data centre 2 (one week) Figure 4.6 Monitored total power demand of data centre 3 (one week)...71 xi

13 Figure 4.7 Monitored total power demand of data centre 4 (one week) Figure 4.8 Monitored total power demand of data centre 5 (one week) Figure 4.9 Monitored total power demand of data centre 6 (one week) Figure 4.10 Total data centre power demand density Figure 4.11 Peak power demand density of total data centre..86 Figure 4.12 Balance between peak and average power demand of data centre as percentage of average power demand Figure 4.13 Total annual energy consumption of data centre, total energy consumption of building 3 is not available Figure 4.14 Over-designing of total data centre power demand..89 Figure 4.15 End-use breakdown of data centre energy consumption 91 Figure 4.16 Energy consumption of IT equipment as percentage of total energy consumption...92 Figure 4.17 Energy consumption of HVAC as percentage of total energy consumption...93 Figure 4.18 Energy consumption of UPS system as percentage of total energy consumption...93 Figure 4.19 Monitored IT equipment power demand of data centre 5.94 Figure 4.20 Actual IT equipment power density..95 Figure 4.21 IT equipment footprint power density Figure 4.22 Over- designing of IT equipment power demand Figure 4.23 UPS systems of data centres Figure 4.24 Profile of UPS power input and output Figure 4.25 UPS efficiency as a function of UPS load factor.101 Figure 4.26 Effect of overall UPS load factor on overall UPS efficiency.102 Figure 4.27 Example of improved UPS efficiency Figure 4.28 Monitored indoor environmental conditions of data centre xii

14 Figure 4.29 HVAC power demand (data centre 4) as a function of outdoor dry-bulb temperature. 107 Figure 4.30 HVAC system power demand density.108 Figure 4.31 HVAC system energy consumption Figure 4.32 Relative energy effectiveness of HVAC system Figure 4.33a Monitored power demand of lighting system of data centre Figure 4.33b Power demand of lighting system of data centre Figure 4.34 Benchmark of total data centre power density Figure 4.35 Benchmark of IT equipment power density Figure 4.36 Benchmark of designed IT equipment power density Figure 4.37 Benchmark of HVAC system power density Figure 4.38 Benchmark of IT equipment power demand/ total data centre power demand Figure 4.39 Benchmark of relative energy effectiveness of HVAC system..121 Figure 4.40 Benchmark of over-designing of IT equipment power demand 122 Figure 5.1 Effect of over-designing of IT equipment power demand on IT equipment energy consumption 126 Figure 5.2 Effect of data centre gross floor area on IT equipment energy consumption 127 Figure 5.3 Effect of IT equipment footprint as percentage of gross area on IT equipment energy consumption.128 Figure 5.4 Effect of IT equipment energy consumption on UPS energy loss Figure 5.5 Effect of overall UPS load factor on overall UPS efficiency Figure 5.6 Effect of IT equipment energy consumption on HVAC system energy consumption Figure 5.7 Effect of gross floor area on HVAC system energy consumption..132 Figure 5.8 Effect of IT equipment footprint area as percentage of gross floor area on HVAC system energy consumption.133 Figure 5.9 Effect of indoor temperature on HVAC system energy consumption xiii

15 Figure 5.10 Effect of HVAC over-designing on HVAC system energy consumption Figure 5.11a Effect of HVAC system over-designing on the relative energy effectiveness of HVAC system Figure 5.11b Effect of HVAC system over-designing on the relative energy effectiveness of HVAC system, after removing data centre Figure 5.12 Effect of over-designing of IT equipment power demand on HVAC system over-designing Figure 5.13 Effect of lighting operation schedule on lighting system energy consumption Figure 5.14 Averaged energy end-use breakdown of data centres studied 139 Figure 5.15 Effect of HVAC system energy consumption on total data centre energy consumption Figure 5.16 Effect of IT equipment energy consumption on total data centre energy consumption Figure 5.17 Effect of over-designing of total data centre power demand on total data centre energy consumption Figure 5.18 Effect of IT equipment footprint area as a percentage of gross floor area on total data centre energy consumption..142 Figure 5.19 Effect of relative energy effectiveness of HVAC system on IT equipment energy consumption as percentage of total data centre energy consumption Figure 5.20 Effect of over-designing of HVAC system on IT equipment energy consumption as percentage of total data centre energy consumption Figure 5.21 Effect of over-designing of IT equipment power demand on over-designing of total data centre power demand..144 Figure 6.1 Typical structure of rule-based fuzzy logic model Figure 6.2 Neural network schematic diagram Figure 6.3 Basic structure of fuzzy logic model 154 Figure 6.4 Standard membership function types Figure 6.5 Fuzzification of variable Figure 6.6 Operator of input aggregation xiv

16 Figure 6.7 Definition of fuzzy rules Figure 6.8 Deffuzification with Centre-of-Area.159 Figure 6.9 Deffuzification with Centre-of-Maximum Figure 6.10 Configuration of neural training.161 Figure 6.11 Training of output variable.162 Figure 6.12 Systematic diagram of modeling procedure Figure 6.13 Model of overall data centre energy performance..165 Figure 6.14 Model of IT equipment energy performance..166 Figure 6.15 Model of HVAC system energy performance 167 Figure 6.16 Integrated model of total data centre energy performance.167 Figure 6.17 Training result in 3-D plot..170 Figure 6.18 Remote Control Unit of neurofuzzy model in Excel..171 Figure 6.19 Deviation between predicted and actual HVAC energy consumptions per unit of gross floor area 174 Figure 6.20 Deviation between predicted and actual IT energy consumption per unit of gross floor area Figure 6.21 Deviation between predicted and actual total data centre energy consumption per unit of gross floor area 176 Figure 6.22 Deviation between predicted and actual HVAC/ IT equipment (energy consumption)..176 Figure 6.23 Deviation between predicted and actual IT equipment/ total data centre (energy consumption).177 xv

17 Chapter 1: Introduction CHAPTER 1 INTRODUCTION 1.1 Background Data centres, the centralized facilities housing large number of IT equipment to perform various functions of storage, management, processing, and exchange of digital data and information, are essential support for the Information and Communication Technology (ICT), which involves universal adoption of computing equipment and network as tools to provide information-based services in the modern knowledge economy. Such facilities are typically set up for web-hosting, central depository information bases of government organizations and research units, intranet, financial transaction processing, telecommunications, and other activities supporting the growing information-based economies. Due to the rapid developments of IT industry and ever-growing Internet, for the proper functioning of IT equipment, data centre as a facility is required to provide guaranteed reliable power supply, security, cooling and connectivity to the Internet. Uninterruptible Power Supply (UPS) and highly powerful HVAC systems are normally employed to provide continuous and reliable support for data centres. Operating these supporting systems and IT equipment, data centres are typically characterized by significantly higher power demands and energy consumptions than those of typical residential or commercial office spaces [Mitchell-Jackson et al., 2002, 1

18 Chapter 1: Introduction 2003; Drecher, 2002; Miller, 2001; LBNL, 2003]. However, the hasty installations, absence of design standards and guidelines, few practical benchmarking and the lack of financial incentives of using energy efficient technologies financial often lead to poor design and inefficient energy use in data centres [E-Source, 2001; Lewis, 2001; Callsen, 2001; Robertson & Romm, 2002]. Over-designed power demands, over-sizing of HVAC and UPS systems, excessive gross floor area, poor layout of IT equipment and supporting systems, unreasonable environmental conditions are universal problems of data centre facilities. In the past few years, the data centre industry had placed increasing power demands on utilities. In April 2001, Consolidation Edison of New York warned [Anderson, 2001] that the plans for 46 such data centres proposed for New York and Westchester Country over a four years period, could impose another 500 MW of demand on its already strained power system. In California, Pacific Gas and Electric (PG&E) claimed [PG&E, 2000] that the power demand from data centres was 341 MW in 2000, and required an additional 1000 MW of power by 2003, which was the equivalent of approximately three new power plants. The energy crisis in California in 2000 and 2001, and the recent historically largest power blackout in the Northern US and Canada have led to rising concerns in society on energy related issues. Data centres characterized with huge energy consumption are inevitably affected. Although the demand of power by data centres has eased somewhat owing to the burst Internet bubble and economy downturn, this is considered a temporary condition. With the expected increase in use of computers worldwide, it is expected that power demand of data centres is set to increase in the 2

19 Chapter 1: Introduction long run. 1.2 Objectives of the study Recently, a number of studies have been conducted for the better understanding of data centre energy performance in the Western countries, such as the US and some European counties. In this case, the term energy performance is a comprehensive evaluation of energy use in data centres, which include not only the energy consumptions and power demands, but also operational energy efficiencies of the supporting systems such as HVAC system and uninterruptible power supply (UPS) units. However, to date in Singapore and other countries in the Southeast Asian region under tropical climate, relevant studies are very limited. With the rapid growth of ICT in this region, data centres will play significantly critical roles in the very close future. In addition, due to the differences of climate and business types of data centres between this region and the Western countries, designers may need more robust benchmark database of local data centres to make better designs and constructions of future data centres. Hence, this study aimed at determining an empirical energy usage pattern of data centres in Singapore, is an exploratory and valuable effort. The main objectives of this study are described as follows: a. To determine an empirical energy usage pattern of data centres in Singapore under tropical weather conditions Energy consumption, power demand and energy efficiency of energy consuming systems and equipment of data centres are to be investigated. Physical parameters 3

20 Chapter 1: Introduction including space usage and indoor environmental conditions which affect energy use of data centres will also be examined. Potential opportunities of improving energy performance in data centres are going to be explored. In addition, practical methods of measurement and evaluation of data centre energy performance are to be developed. b. To develop whole-system based models of data centre energy performance To examine data centre energy performance and potential energy savings from a whole-system point of view, models that incorporate energy usages of various energy consuming systems and other affective variables are to be established. 1.3 Scope and limits Case studies and field measurements are conducted in the six data centres in Singapore. The variables indicating or affecting the energy performance of data centres are examined. These include the energy consumptions, power demands, and operational energy efficiencies of various energy consuming systems, as well as the space use, indoor environmental conditions, strategies of design and operation of data centres. Based on the data obtained, initial benchmarking of energy performance is conducted among the selected data centres. The result of this study is compared with the comparative studies conducted by Lawrence Berkeley National Laboratory (LBNL) in the United States. The reasons for inefficient energy use in the design and operation stages of data centres are discussed. Recommendations for improving energy performance of data centres are provided. For the purpose of accomplishing an integrative design and performance evaluation of 4

21 Chapter 1: Introduction data centres, neuro-fuzzy network modeling is employed in this study. Models are developed based on the results of case studies and field measurements. Using the whole-system based concept and incorporating energy performance of energy consuming systems and the various affecting factors, the models developed can aid in evaluating energy performance and predicting energy saving potentials of data centres. However, this study is faced with certain limitations as follows: a. Constraints of time, resources and access to data centres lead to a small sample size of data centres. Six data centres are involved in this study. b. Information obtained from the data centre operators may not be truly accurate. In addition, information such as area, power capacity and specifications of IT equipment is not always readily available, owing to the fact that the data centres are one of the most confidential areas in buildings. c. The bills of data centres electricity energy consumption, separated from other parts of the building are often not available. As a result, there are no bill calibrations on the measured energy data of data centres, so that the data is inevitably subjected to a degree of error. d. Due to the constraint of measuring instruments, some of the key parameters, such as cooling load of data centres and temperatures of supply and return chill water, are not determined. 5

22 Chapter 1: Introduction 1.4 Outline of the thesis Chapter 1 Introduction In this chapter, the background of data centres, previous studies associated with energy concerns over data centres, research objectives, scope and limits of the study are described. Chapter 2 Review of literature on data centre energy performance This chapter provides a detailed review of the data centre industry and the studies of energy performance of these high energy consuming facilities. The issues highlighted include the definitions and classifications of data centres, common designs, profiles of energy consumptions and power demands, reasons for over-designing and inefficient energy use of data centres, indicators of data centre energy performance, as well as the future trends of data centres industry and their potential impact on the design of data centres. Chapter 3 Methodology and research design This chapter outlines the research methodology and research design. These include a flow chart illustrating the research design, selection of the data centres, modeling of data centre energy performance, methods of case studies and field measurements, instrumentations, calculating methods of energy consumption and power demand, uncertainty analysis of data measured, benchmarking of data centre energy performance, and the whole-system based neuro-fuzzy network modeling. 6

23 Chapter 1: Introduction Chapter 4 Measurements and results In this chapter, the results of the case studies and filed measurement, and the energy performance benchmarking of data centres are presented and discussed. Reasons of inefficient energy use and improper designs of data centres are discussed. Recommendations for the improving the design and operation of future data centres are provided. Chapter 5 Parameter analysis of data centre energy performance In this chapter, based on the data obtained from field measurements and preliminary models of various energy consuming systems, parameter analysis is developed for the better understanding of data centre energy performance. Chapter 6 Neuro-fuzzy network modeling of data centre energy performance This chapter describes an exploratory study of the whole-system based modeling of data centre energy performance using neuro-fuzzy network approach. The rationale behind adopting neuro-fuzzy network modeling to the study of data centre energy performance is discussed. The results and potential contributions of the models are presented and discussed. Chapter 7 Conclusions This chapter concludes the study with a summary of the main findings, contributions of the study and the recommendations for future studies. 7

24 Chapter 2: Review of Literature on Data Centre Energy Performance CHAPTER 2 REVIEW OF LITERATURE ON DATA CENTRE ENERGY PERFORMANCE 2.1 Introduction This chapter presents the concerns of society on the energy performance of data centres and main findings of the previous studies. These mainly include the definitions and classifications of data centres, common designs, actual energy consumption and power demand, over-designing and inefficient energy use of various energy consuming systems, environmental conditions, opportunities of energy saving, and difficulties of predicting power demand for the future data centres, as well as the alternative indicators of data centre energy performance. 2.2 Data centre and its classification To date, a number of the definitions of data centre are around. Robertson and Romm (2002) define that data centres house the computers, servers, switches, routers, data storage devices and related equipment used to operate the digital economy; they can be relatively small facilities owned by a single company and dedicated to processing its data alone, while they also can be huge facilities where many companies can rent floor space containing wire cages filled with racks to house their servers and other equipment or the facilities having a single owner who provides managed information technology (IT) services and acts as an application services provider (ASP) on an out-sourcing basis. 8

25 Chapter 2: Review of Literature on Data Centre Energy Performance Beck (2001) defines that data centre is a generic term used to describe a range of facilities that use high concentrations of computers, servers, and digital electronic equipment to host website, store, process and transfer electronic data, and otherwise support a digital electronic economy. Beck states that there are many flavors of data centres, each of which may have its own specific issues related to power demand and load profiles. A few common terms and client profiles for data centres are: Data storage and Internet hosting facilities, also known as server farms or Internet hotels that perform a variety of functions. Internet service provider routers (ISPs), dedicated specifically to supporting the Internet. Telecommunication switches, known as telecoms or telcos. These are more energy demanding than typical Internet data centres [Uptime, 2000]. Managed data centres, where racks and computer equipment are owned by data centre owner, but leased to tenants. Co-located server hosting facilities, also known as CoLos, where rack space is leased by tenants and computer equipment is owned and operated by tenants. As tenants may move in and out, upgrade their computers frequently, and have a disconnection between the energy-using facility and the billing department, energy demands tend to have greater fluctuations and to be less well characterized than corporate data centres. 9

26 Chapter 2: Review of Literature on Data Centre Energy Performance Corporate data centres, where racks and computer equipment are wholly owned and operated by the corporation. As a sole entity owns and operates this type of data centre, energy demand may be better characterized than other data centres. Other examples of this type of data centres may be found in universities, banks, and governmental organizations. In the report Energy- and Eco-Efficiency of Data Centres, Dr. Aebischer (2003) from Switzerland describes data centre as a building or parts of a building (a hall or a room) accommodating servers and other ICT equipment. The function of the ICT equipment hosted in a data centre is manifold. It may be an element of an intranet, part of the public infrastructure of the Internet and the telecommunication-system or a mixture of these two functions. Aebischer also says that there are many ways to classify data centres, e.g. regarding the service provided: ISP = internet service provider, ASP = application service provider, FSP = full service provider, Or with respect to the ownership of the data centre and of the ICT equipment: Corporate data centres Collocation data centres Managed data centres 10

27 Chapter 2: Review of Literature on Data Centre Energy Performance The more updated definition of data centre is given by the researchers from Lawrence Berkley National Laboratory (LBNL) (2003). They define data centre as a special facility that performs one or more of the following functions: Store, manage, process, and exchange digital data and information; Provide application services or management for various data processing, such as web hosting internet, intranet, tele-communication and information technology. In addition, spaces that primarily house office computers including individual servers associated with work stations as data centres. Other terms that refer to such facilities that meet the definition of data centre, include computer centre, data storage and hosting facility, server farm, data farm, data warehouse, collocation facility, co-located server hosting facility (CoLo), corporate data centre, managed data centres, internet hotel, internet service provider (ISP), application service provider (ASP), full service provider (FSP), wireless application service provider (WASP), telecommunication hotel (or telco hotel), carrier hotel, internet hotel, telecommunications carriers, or other data networks. Based on the review of previous studies, in this study, data centre is defined as a generic term used to describe the centralized facilities housing large number of IT equipment to perform various functions of storage, management, processing, and exchange of digital data and information, and being the essential support for Information and Communication Technology (ICT), which involves universal 11

28 Chapter 2: Review of Literature on Data Centre Energy Performance adoption of computing equipment and network as tools to provide information-based services in the modern knowledge economy. Such facilities are typically set up for web-hosting, central depository information bases of government organizations and research units, Intranets, financial transaction processing, telecommunications, and other activities that support the growing information-based economies. In Singapore, governmental organizations, universities, banks, and many companies often have data centres or computer rooms of their own in the commercial or administrative buildings. This type of facility is categorized as corporate data centre, where the data centre room and the equipment inside are wholly owned and operated by a single corporation. In these buildings, data centres provide local area network (LAN), external internet access, printing, data storage and processing, and other information services for office use. This study draws focus on the energy performance of the corporate data centres. 2.3 Data centre common designs For the proper functioning of IT equipment, data centre as a facility is required to provide guaranteed reliable power supply, security, cooling and connectivity to the Internet 24 hours a day and 7 days a week. Although there are various names, business models and sizes of data centres, such facilities all share certain common attributes Typical energy consuming systems Data centres commonly employ utility power as the main power, UPS as the power conditioner, batteries and diesel generator as the back-up power recourses. In case of a power failure, batteries will act as the back-up power resource of critical IT equipment 12

29 Chapter 2: Review of Literature on Data Centre Energy Performance for about 15 to 30 minutes. The diesel generator will then be started to generate power till the main power come back. In some data centres apart from the main UPS, sub-ups systems are employed to provide extra protection for highly critical servers or storage devices. During power failures, the back-up power would give personnel ample time to save data and shut down the IT equipment properly. Data centres contain concentrated IT equipment, which commonly includes servers, data storing devices, network devices, monitors, and others. As the critical load in data centres, IT equipment is typically powered by uninterruptible power supply (UPS). Non-critical loads that do not connect to UPS such as PCs, printers, fax machines, copiers, are not considered as IT equipment and classified as office equipment in this study. HVAC system is the major energy consuming system in a data centre. In this study, the HVAC system mainly refers to the air-conditioning system for space cooling. CRAC that stands for computer room air conditioner is the typical air-conditioning system used in data centres. Commonly employed CRAC systems include the water-cooled unit, air-cooled unit, and air conditioning unit using the chilled water supplied by central chiller plant. In some cases, central air-conditioning and FCU systems are also adopted. Another energy consuming component is the UPS system. Electricity is wasted inside the UPS during the process of charging battery and current transform. The remaining energy consuming systems include lighting and other auxiliary systems 13

30 Chapter 2: Review of Literature on Data Centre Energy Performance such as current transformer, back-up diesel generator, building control and automation system, fire suppression system, office equipment, and telecommunication system. Appendix A lists the common energy consuming systems of data centres. Figure 2.1 illustrates the energy consumption distribution in a typical data centre. HVAC Utility Distribution Lighting UPS IT equipment Back-up diesel generator Others (Such as office equipment, building control system and fire suppression system) Figure 2.1 Energy consumption distribution in a typical data centre Infrastructure and space use Data centres are typically built based on raised floors. Although not necessarily, raised floor provides flexibility in electrical and network cabling, as well as air conditioning. The purpose of a raised floor is to channel cold air from the HVAC units and direct it up where it is needed to cool equipment. It also acts as an out-of the-way area to route network and power cables, and a framework for equipment grounding. Studies [Mitchell-Jackson, 2001; Beck, 2001; Snevely, 2002] have generally shown that based on the raised floor area, only 25-40% is occupied by IT equipment, aisle and 14

31 Chapter 2: Review of Literature on Data Centre Energy Performance other necessary open space constitutes 40-50% of the floor area, and the remaining 10-35% is for supporting systems such as HVAC, UPS and office equipment Occupants Normally, very few people work in data centres. Unless the office or control room is located together in data centre area, personnel would only enter data centre room when data-processing or equipment maintenance is needed Environmental conditions Temperature is crucial for the proper operation of data centres. This is because electrical energy consumed by IT equipment is finally transformed to heat; however digital hardware is unable to bear the high temperature. The more power a CPU draws, the hotter it gets. The hotter a CPU gets, the more likely it will fail, and likely, cause other components to fail. said Dr. Wu-chun Feng (2003). More concretely, the Arrenhius equation (when applied to microelectronics) predicts that the failure rate of a given system doubles with every 10 o C increase in temperature. In the design guideline Enterprise Data Centre Design and Methodology, it is recommended by Sun Microsystems [Snevely, 2002] that an ambient dry-bulb temperature range of 21 to 23 o C is optimal for system reliability and operator comfort in data centres. It is claimed that most digital hardware can operate within a wider psychometric range, but a temperature level near 22 o C is the best, for it is easier to maintain safety associated with relative humidity levels; another reason for keeping the room temperature as close to the optimal temperature as possible is to give the greatest 15

32 Chapter 2: Review of Literature on Data Centre Energy Performance buffer against problems as activities can change the temperature profile. Sun Microsystems also recommends that ambient relative humidity (RH) level maintained at approximately 50%, plus or minus 5% could be the optimal range in data centres. Most data processing equipment can operate within a fairly wide RH range of 20% to 80%, but maintaining RH near the optimal level can prevent corrosion, electrostatic discharge (ESD) and provide operating time buffer in case of a system failure. As a result, the reliability and the life expectancy of the data centre equipment can be enhanced. It should be noted that discussion of wider ranges of humidity and temperature control which can be tolerated by electronics equipment is raised recently. Were humidity levels allowed to fluctuate more, energy consumed to achieve humidity control could be saved, and if even small increases in ambient temperature are acceptable, significant energy saving will result. ASHRAE has established a new Technical Commettee (TC 9.9) [TC 9.9, 2004] to focus on High Density Electronic Equipment Facility Cooling. Environmental specifications for various classes of electronic equipment and standardized methods of determining temperature in data centres are focused in the guideline which is being developed Power quality and reliability Electronic equipment operates with continuous flow of electrons. Tiny disruptions in the current or voltage of the electric power supply will corrupt data or crash computers. For data centres, these fluctuations can cause large financial losses. For this reason, data centres are required to provide high availability and quality of power 24 hours a 16

33 Chapter 2: Review of Literature on Data Centre Energy Performance day, 7 days a week. To ensure the power quality and reliability, data centres traditionally rely on electric grid plus power conditioning equipment and back-up power supplies. Data centre owners normally install N+1 or 2N redundant power system composite of dual utility feeder from separate substations connected to on-site power conditioning equipment and uninterruptible power supply (UPS), where the raw power is conditioned to be more steady, and then supplied to the IT equipment. The back-up power supplies such as batteries, flywheel storage, ultra-capacitors, and on-site power source such as diesel generators, will take over the power supply of the critical loads in case of a power failure, until the utility power returns. In addition, the HVAC systems also need to be running continuously, because the heat load generated by the IT equipment is so intense that they will subsequently over-heat themselves and malfunction, if there is no cooling. For this reason, power redundancies on HVAC system are also necessary. To provide higher power reliability and improve the energy efficiency of data centres, applications of distributed generation (DG) and combined heat and power (CHP) are believed to be the best solution for those mission-critical facilities [Beck, 2001; Mitchell-Jackson et al., 2002; Plant-TECH Associates, 2002]. For example, use of fuel cells with grid as backup power, could simplify power conversions, be a ready source of uninterruptible power and could eliminate costly and inefficient equipment. However, these are already beyond the scope of this study. 17

34 Chapter 2: Review of Literature on Data Centre Energy Performance 2.4 Data centre energy performance Data centres put substantial burden on utilities Prior to the dot.com crash and 2001 economic slowdown, the burgeoning data centre industry placed increasing power demands on utilities. In the US, Seattle s Puget Sound Energy (PSE) reported [Cook, 2000; Grahame and Kathan, 2001] that in August 2000, there were requests for 445 MW of power from data centres which are a small area near Southcentre mall. This power draw is equivalent to that of six oil refineries. By September that year, requests had reached 700MW, briefly rising to over 1,000 MW until the dot.com bust. In April 2001, Consolidation Edison of New York warned [Anderson, 2001] that the plans for 46 data centres proposed for New York and Westchester Country over a four years period could impose another 500 MW of demand on its already strained power system. In California, Pacific Gas and Electric (PG&E) claimed [PG&E, 2000] that the power demand from data centres was 341 MW in 2000, and requires an additional 1000 MW of power by 2003 the equivalent of approximately three new power plants. Back at San Jose, California in March 2001, city officials gave preliminary approval to what would be the world s largest server farm, which would occupy 10 buildings on more than 170 acres in the city s Alviso area, with an aggregate area of 204,000 square meter, and a total projected energy use 180 MW by 2005 [Lazarus, 2001]. The breakout of energy crisis in California on 2000 and 2001, and the recent historically largest power blackout in the Northern US and Canada have raised the concerns of sustainable energy use. Data centres characterized with huge energy consumption are surely involved in this energy issue. Although the demand for power 18