Technical ReportPDF Available

United States Data Center Energy Usage Report

Authors:

Abstract and Figures

This report estimates historical data center electricity consumption back to 2000, relying on previous studies and historical shipment data, and forecasts consumption out to 2020 based on new trends and the most recent data available. In 2014, data centers in the U.S. consumed an estimated 70 billion kWh, representing about 1.8% of total U.S. electricity consumption. This report shows that data center electricity consumption increased by about 4% from 2010-2014, a large shift from the 24% percent increase estimated from 2005-2010 and the nearly 90% increase estimated from 2000-2005. Energy use is expected to continue slightly increasing in the near future, increasing 4% from 2014-2020, the same rate as the past five years. Based on current trend estimates, U.S. data centers are projected to consume approximately 73 billion kWh in 2020. A combination of efficiency trends has resulted in a relatively steady U.S data center electricity demand over the past 5 years, with little growth expected for the remainder of this decade. Along with the energy efficiency resource already achieved, there are additional energy efficiency strategies and technologies that could significantly reduce data center electricity use below the approximately 73 billion kWh demand projected in 2020. Many of these efficiency strategies are already successfully employed in some data centers while others are emerging technologies that will be commercially available in the near future. The potential impact from an adoption of additional energy efficiency strategies is explored, which estimate an annual saving in 2020 up to 33 billion kWh, representing a 45% reduction in electricity demand when compared to current efficiency trends. Full text is available at: http://eta.lbl.gov/publications/united-states-data-center-energy-usag
Content may be subject to copyright.
ERNEST ORLANDO LAWRENCE
BERKELEY NATIONAL LABORATORY
LBNL-1005775
United States Data Center
Energy Usage Report
Arman Shehabi, Sarah Smith, Dale Sartor, Richard Brown, Magnus Herrlin
Environmental and Energy Impact Division, Lawrence Berkeley National
Laboratory
Jonathan Koomey
Steyer-Taylor Center for Energy Policy and Finance, Stanford University
Eric Masanet
McCormick School of Engineering, Northwestern University
Nathaniel Horner, Inês Azevedo
Climate and Energy Decision Making Center, Carnegie Mellon University
William Lintner
Federal Energy Management Program, U.S. Department of Energy
June 2016
This work was supported by the Federal Energy Management Program of the
U.S. Department of Energy under Lawrence Berkeley National Laboratory Contract
No. DE-AC02-05CH1131
Disclaimer
This document was prepared as an account of work sponsored by the United States
Government. While this document is believed to contain correct information, neither the United
States Government nor any agency thereof, nor The Regents of the University of California, nor
any of their employees, makes any warranty, express or implied, or assumes any legal
responsibility for the accuracy, completeness, or usefulness of any information, apparatus,
product, or process disclosed, or represents that its use would not infringe privately owned
rights. Reference herein to any specific commercial product, process, or service by its trade
name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply its
endorsement, recommendation, or favoring by the United States Government or any agency
thereof, or The Regents of the University of California. The views and opinions of authors
expressed herein do not necessarily state or reflect those of the United States Government or
any agency thereof, or The Regents of the University of California.
Ernest Orlando Lawrence Berkeley National Laboratory is an equal opportunity employer.
i
Acknowledgments
The Green Grid Association provided comments to draft versions of this report that were
contributed by their members representing major data center networking equipment
manufacturers, server and storage equipment manufacturers, major software providers, and
large data center end users/owners.
Input data for this report was provided by the Green Grid Association, Data Center Dynamics,
the IT Industry Council, and the International Data Corporation.
We would also like to acknowledge and thank the expert reviewers from approximately 30
organizations that took the time to answer questions from the authors and provided comments
on draft versions of the report.
The research reported in this report was conducted by Lawrence Berkeley National Laboratory
with support from the Department of Energy Federal Energy Management Program. Lawrence
Berkeley National Laboratory is supported by the Office of Science of the United States
Department of Energy and operated under Contract Grant No. DE-AC02-05CH11231.
Citation
Shehabi, A., Smith, S.J., Horner, N., Azevedo, I., Brown, R., Koomey, J., Masanet, E., Sartor,
D., Herrlin, M., Lintner, W. 2016. United States Data Center Energy Usage Report. Lawrence
Berkeley National Laboratory, Berkeley, California. LBNL-1005775
ii
Table of Contents
1 Introduction .......................................................................................................................... 1
1.1 Report Scope .................................................................................................................. 1
1.2 Report Organization ....................................................................................................... 2
2 Estimates of U.S. Server and Data Center Energy Use .................................................... 3
2.1 Methodology Overview ................................................................................................... 3
2.2 Data ................................................................................................................................ 5
2.2.1 IDC Worldwide Trackers ............................................................................................. 5
2.2.2 SPEC Database .......................................................................................................... 5
2.2.3 SERT Database .......................................................................................................... 5
2.3 IT and Infrastructure Equipment Estimates .................................................................... 6
2.3.1 Server Energy Use ...................................................................................................... 6
2.3.2 Storage Energy Use .................................................................................................. 14
2.3.3 Network Energy Use ................................................................................................. 17
2.3.4 Data center space type classifications ...................................................................... 19
2.3.5 Data center energy consumption .............................................................................. 24
2.3.6 Data center water consumption ................................................................................ 28
3 Energy Use Associated with Federal Government Servers and Data Centers ............ 29
4 Expected Energy Savings Opportunities ........................................................................ 31
4.1 Energy efficiency trends ............................................................................................... 31
4.2 Improved Management scenario .................................................................................. 31
4.3 Best practices scenario ................................................................................................ 32
4.4 Hyperscale Shift scenario ............................................................................................. 36
4.5 Scenario results ............................................................................................................ 37
5 Indirect energy impacts .................................................................................................... 41
5.1 Energy impact taxonomy .............................................................................................. 42
5.2 Energy impact estimation ............................................................................................. 45
5.3 Pathway forward ........................................................................................................... 45
6 Future Work ........................................................................................................................ 46
6.1 Server utilization and power proportionality .................................................................. 46
6.2 Workload variation ........................................................................................................ 47
6.3 Barriers to hyperscale shift ........................................................................................... 47
6.4 Beyond PUE ................................................................................................................. 47
6.5 Beyond 2020 ................................................................................................................ 47
iii
Table of Figures
Figure 1. Volume Server Supply Chain ......................................................................................... 4
Figure 2. Equipment Types Modeled in Energy Estimation .......................................................... 4
Figure 3. Schematic of Modeling Approach .................................................................................. 5
Figure 4. Unbranded Server Installed Base and Underlying Assumptions ................................... 7
Figure 5. Total Volume Server Installed Base Estimates from Three Studies .............................. 8
Figure 6. Volume Server Installed Base 2000-2020 Disaggregated by Processor Count and
Vendor Type ................................................................................................................................. 9
Figure 7. Average Power Draw Assumptions for Mid-Range and High-End Servers ................. 10
Figure 8. Assumed Dynamic Range of Volume Servers ............................................................. 12
Figure 9. Dynamic Range of 1- and 2-Socket Servers in SPEC Database................................. 12
Figure 10. Maximum and Effective Average Power Estimates for Volume Servers ................... 13
Figure 11. Total U.S. Annual Direct Server Electricity Consumption by Server Class ................ 13
Figure 12. Total U.S. Data center Storage Installed Base in Capacity (TB) ............................... 14
Figure 13. Estimated Average Capacity of U.S. Data center Storage Drives ............................. 15
Figure 14. Total U.S. Data Center Storage Installed Base in Drive Count.................................. 15
Figure 15. Average Wattage of Storage Drives in U.S. Data Centers ........................................ 16
Figure 16. Total U.S. Data Center Storage Electricity Consumption .......................................... 17
Figure 17. Total U.S. Data center Installed Base of Network Ports ............................................ 18
Figure 18. Assumed Network Power for Four Port Speeds ........................................................ 18
Figure 19. Total U.S. Data center Network Equipment Electricity Consumption ........................ 19
Figure 20. Total Server Installed Base by Data center Space Category .................................... 23
Figure 21. Total Electricity Consumption by Technology Type ................................................... 25
Figure 22. Total Electricity Consumption by Space Type ........................................................... 26
Figure 23. Historical Data Center Total Electricity Use ............................................................... 26
Figure 24. Data Center Electricity Consumption in Current Trends and 2010 Energy Efficiency
Scenarios .................................................................................................................................... 28
Figure 25. Direct vs. Indirect U.S. Data Center Water Consumption .......................................... 29
Figure 26. Total U.S. Data Center Water Consumption by Space Type ..................................... 29
Figure 27. Average Volume Server Dynamic Range for Current Trends and Best Practices
Scenarios .................................................................................................................................... 33
Figure 28. Volume Server Installed Base for Current Trends and Best Practices Scenarios ..... 35
Figure 29. Network Installed Base for 10 GB and 40 GB ports in Current Trends and Best
Practices Scenarios .................................................................................................................... 35
Figure 30. Storage Disk Power Consumption for Current Trends and Best Practices Scenarios
.................................................................................................................................................... 36
Figure 31. Volume Server Installed Base in CT and HS Scenarios ............................................ 37
Figure 32. Server Electricity Use for All Scenarios ..................................................................... 38
Figure 33. Network Electricity Use for Current Trends and Best Practices Scenarios. ............... 38
Figure 34. Infrastructure Electricity Use for All Scenarios ........................................................... 39
Figure 35. Total Electricity Consumption for All Scenarios ......................................................... 40
Figure 36. Water Consumption for All Scenarios ........................................................................ 40
Figure 37. Taxonomy of Energy Effects from Adoption of ICT, from Horner et al....................... 44
iv
Table of Tables
Table 1. Average Active Volume Server Utilization Assumptions ............................................... 10
Table 2. Typical IT Equipment and Site Infrastructure System Characteristics by Space Type . 21
Table 3. Allocation of Data Center Equipment Across Space Types .......................................... 22
Table 4. 2014 PUE by Space Type ............................................................................................. 24
Table 5. Servers in Federal Data centers Tracked by OMB ....................................................... 30
Table 6. PUE and Redundancy Values for Efficiency Scenarios ................................................ 32
Table 7. Best Practices Scenario Consolidation Parameters ..................................................... 33
Table 8. Taxonomy of ICT Energy Effects from Horner et al. ..................................................... 43
Version Updates
6/27/16: Original report posted
7/12/16: Equation 3 updated to read “m*(1-u)” instead of “m*u”
ES-1
Executive Summary
This report estimates historical data center electricity consumption back to 2000, relying on
previous studies and historical shipment data, and forecasts consumption out to 2020 based on
new trends and the most recent data available. Figure ES-1 provides an estimate of total U.S.
data center electricity use (servers, storage, network equipment, and infrastructure) from 2000-
2020. In 2014, data centers in the U.S. consumed an estimated 70 billion kWh, representing
about 1.8% of total U.S. electricity consumption. Current study results show data center
electricity consumption increased by about 4% from 2010-2014, a large shift from the 24%
percent increase estimated from 2005-2010 and the nearly 90% increase estimated from 2000-
2005. Energy use is expected to continue slightly increasing in the near future, increasing 4%
from 2014-2020, the same rate as the past five years. Based on current trend estimates, U.S.
data centers are projected to consume approximately 73 billion kWh in 2020.
Many factors contribute to the overall energy trends found in this report, though the most
conspicuous change may be the reduced growth in the number of servers operating in data
centers. While shipments of new servers into data centers continue to grow every year, the
growth rate has diminished over the past 15 years. From 2000-2005, server shipments
increased by 15% each year resulting in a near doubling of servers operating in data centers.
From 2005-2010, the annual shipment increase fell to 5%, partially driven by a conspicuous
drop in 2009 shipments (most likely from the economic recession), as well as from the
emergence of server virtualization across that 5-year period. The annual growth in server
shipments further dropped after 2010 to 3% and that growth rate is now expected to continue
through 2020. This 3% annual growth rate coincides with the rise in very large “hyperscale” data
centers and an increased popularity of moving previously localized data center activity to
colocation or cloud facilities. In fact, nearly all server shipment growth since 2010 occurred in
servers destined for large hyperscale data centers, where servers are often configured for
maximum productivity and operated at high utilization rates, resulting in fewer servers needed in
the hyperscale data centers than would be required to provide the same services in traditional,
smaller, data centers.
Along with total server count, the power demand for each server has also changed. While
server power requirements were observed to be increasing from 2000-2005, power demand
appears to have stayed fairly constant since 2005. Additionally, servers are improving in their
power scaling abilities, thus reducing power draw during idle periods or when at low utilization.
Efficiency improvements in storage, network and infrastructure also influence the electricity
estimates in this report. Storage devices are becoming more efficient on a per-drive basis, with
the growth in drive storage capacity projected to outpace increases in data storage demand by
2020, ultimately reducing the number of physical drives needed throughout data centers.
Recent estimates of network port power consumption are now much lower than estimates from
the past decade. Increased awareness in data center infrastructure operations (e.g. cooling) has
resulted in improved efficiency across data center types, though the most significant
infrastructure impact observed in this report is the recent growth in hyperscale data centers that
are often innovatively designed to maximum infrastructure efficiency.
ES-2
The combination of these efficiency trends has resulted in a relatively steady U.S data center
electricity demand over the past 5 years, with little growth expected for the remainder of this
decade. It is important to note that this near constant electricity demand across the decade is
occurring while simultaneously meeting a drastic increase in demand for data center services;
data center electricity use would be significantly higher without these energy efficiency
improvements. A counterfactual scenario was created for this study that estimates what data
center energy consumption would have been if industry energy-savings efforts were halted in
2010. For this scenario, the follow metrics remain static at 2010 industry-wide levels from 2010-
2020:
Average server utilization
Server power scaling at low utilization
Average power draw of hard disk drives
Average power draw of network ports
Average infrastructure efficiency (i.e., PUE)
The resulting electricity demand, shown in Figure ES-1, indicates that more than 600 additional
billion kWh would have been required across the decade.
Figure ES-1 Projected Data Center Total Electricity Use
Estimates include energy used for servers, storage, network equipment, and infrastructure in all U.S. data
centers. The solid line represents historical estimates from 2000-2014 and the dashed lines represent five
projection scenarios through 2020; Current Trends, Improved Management (IM), Best Practices (BP),
Hyperscale Shift (HS), and the static 2010 Energy Efficiency counterfactual.
ES-3
Note that this scenario does not halt the technological advancements of the computing industry
in terms of performance, and therefore metrics such as computational performance (i.e.,
computations/second per server), the electrical efficiency of computations (i.e., computations
per kWh), storage capacity (i.e., TB per drive), and port speeds (i.e., Gb per port) are all
assumed to progress as normal. See Section 2.3.5 in the main body of this report for more
details regarding the assumptions in this counterfactual scenario.
Along with the considerable energy efficiency resource already achieved, there are additional
energy efficiency strategies and technologies that could significantly reduce data center
electricity use below the approximately 73 billion kWh demand projected in 2020. Many of these
efficiency strategies are already successfully employed in some data centers while others are
emerging technologies that will be commercially available in the near future. Recently observed
efficiency trends are incorporated into a “current trends” scenario. The potential impact from a
more aggressive adoption of the energy efficiency strategies is explored through additional
projections that apply a combination of the three following efficiency scenarios:
The “improved management” scenario includes energy-efficiency improvements beyond
current trends that are either operational or technological changes that require minimal
capital investment. This scenario represents a focus on improving the least efficient
components of the data center stock by employing practices already commonly used in
data centers.
The “best practices” scenario represents the efficiency gains that can be obtained
through the widespread adoption the most efficient technologies and best management
practices applicable to each data center type. This scenario focuses on maximizing the
efficiency of each type of data center facility.
The “hyperscale shift” scenario represents an aggressive shift of data center activity
from smaller data centers to larger data centers. While the current trend scenario
already incorporates some movement towards more server use in large data centers,
this scenario assumes the majority of servers in the remaining small data centers are
also relocated.
In addition to applying each of these scenarios independently, two additional scenarios
demonstrate the combination of a “hyperscale shift” scenario in conjunction with either the
“improved operation” or “best practices” scenario. Figure ES-1 shows that these five scenarios
yield an annual saving in 2020 up to 33 billion kWh, representing a 45% reduction in electricity
demand when compared to current efficiency trends.
1
1 Introduction
Data centers primarily contain electronic equipment used for data processing (servers), data
storage (storage equipment), and communications (network equipment). Collectively, this
equipment processes, stores, and transmits digital information and is known as “information
technology” (IT) equipment. Data centers also usually contain specialized power conversion and
backup equipment to maintain reliable, high-quality, power as well as environmental control
equipment to maintain the proper temperature and humidity for the IT equipment.
As our economy and society continue to shift towards increased digital information
management, data centers have become ubiquitous – they are found in nearly every sector of
the economy – and are essential to the function of communication, business, academic, and
governmental systems. All but the smallest companies have some kind of data center needs,
and larger companies often have tens, or even hundreds, of data centers. Smaller data centers
are commonly located within large commercial buildings, while larger data centers tend to be
buildings constructed specifically for their use that can be up to several hundred thousand
square feet in size. Universities, municipalities, and government institutions also use and
operate data centers for information management and communication functions.
The energy used by the nation’s servers and data centers is significant. In a 2007 Report to
Congress,1 the data center sector was estimated to have consumed about 61 billion kilowatt-
hours (kWh) in 2006 (1.5 percent of total U.S. electricity consumption) for a total electricity cost
of about $4.5 billion (2006 dollars). This estimated level of electricity consumption is similar to
the amount of electricity consumed by approximately 5.8 million average U.S. households. The
electricity use of the nation’s servers and data centers in 2006 was more than double the
electricity that was estimated to have been consumed for this purpose in 2000. The
accompanying methodology article2 to the 2007 Report, published in 2011 with updated
methods and inputs, estimated 2008 data center electricity demand to be 69 billion kWh, 1.8%
of total U.S. electricity sales. Another study published in 20113 revealed that electricity use by
U.S. data centers in 2010 constituted about 2% of overall electricity use in that year, and that
the rate of growth in energy use slowed significantly in the period 2005-2010, compared to
2000-2005. This trend was likely a result of the 2008-09 economic crisis and the increased
adoption of virtualization and other energy efficiency practices in the data center industry.3 This
report provides updated estimates of current data center energy use, updated historical
estimates of energy use back to the year 2000, and projections for energy use through 2020.
1.1 Report Scope
This report builds on previous modeling efforts and updates key inputs to the data center model
used in the 2007 Environmental Protection Agency Report to Congress on Server and Data
Center Efficiency, Public law 109-4311.1 The scope of this report includes updates to the
following sections from the 2007 study:
Trends in Growth and Energy Use Associated with Servers and Data Centers in the U.S.
2
Potential Energy and Cost Savings through Improved Energy Efficiency
Additionally, this report provides the following insights as outlined in the North American Energy
Security and Infrastructure Act of 2015:4
1. A comparison and gap analysis of the estimates and projections contained in the original
report with new data regarding the period from 2008 through 2015;
2. An analysis considering the impact of information technologies, including virtualization
and cloud computing, in the public and private sectors;
3. An evaluation of the impact of the combination of cloud platforms, mobile devices, social
media, and big data on data center energy usage;
4. An evaluation of water usage in data centers and recommendations for reductions in
such water usage; and
5. Updated projections and recommendations for best practices through fiscal year 2020.
1.2 Report Organization
This report serves as an update to the 2007 Environmental Protection Agency Report to
Congress on Server and Data Center Efficiency, Public law 109-43111 (henceforth referred in
this study as the 2007 Report). The 2007 Report provides detailed background information such
as data center equipment layout and cooling configurations, which are therefore not discussed
here. Rather, this report focuses on new trends and data available since the 2007 Report,
methodology of the current estimate, and details of the new results and findings. Chapter 2 of
this report describes the methodology used for the current energy and water use estimates.
Although the current methodology follows the framework presented in the 2007 Report, recent
trends such as the proliferation of “unbranded” servers and storage equipment, also referred to
as “self-assembled,” “whitebox,” or original design manufacturer (ODM) servers require
additional data input and characterization. Unbranded servers are associated with the
increasing number of very large “hyperscale” data centers and often bypass large, branded
server vendors (e.g. Hewlett-Packard, Dell) by being bought directly from the manufacturing
companies that build servers (e.g., Quanta, Wistron, and Foxcom). See Chapter 2 for a more
detailed description of unbranded servers. Chapter 3 of the report is devoted to the discussion
of servers and data centers owned and operated by the federal government. Chapter 4
describes a number of key trends observed in the data center industry, such as server
consolidation, colocation, and the shift to cloud-based platforms. These trends, along with
energy efficiency strategies, are used to generate alternative energy use projection scenarios
out to the year 2020. While rapid growth in the Internet industry has resulted in significant direct
energy use in data centers, it can also indirectly impact certain resources in other sectors and
possibly promote other efficiencies in society that were previously not achievable. These
observations are discussed in Chapter 5. Finally, Chapter 6 provides suggestions for future
research to address challenges that have been identified but not addressed in this report, which
could improve future energy use estimates in such a rapidly evolving industry with little
publically available data.
3
2 Estimates of U.S. Data Center Energy Use
2.1 Methodology Overview
The 2007 Report provided a series of data center energy use projections for 2007-2011
developed for five different efficiency scenarios. Detailed assumptions used for those efficiency
measures in each scenario can be found in Brown et al. (2007),1 while the modeling algorithms
developed for these energy use projections are presented in Masanet et al. (2011).2 Additional
studies were conducted by Koomey3 5 6 to assess the growth in data center electricity use from
2000 to 2010 for the U.S. and the world, using methods similar to the 2007 Report. The 2010
data center electricity use estimated by Koomey (2011) most closely aligns with the “improved
operation scenario” projection in the 2007 Report.
The current study differs from previous work in its categorization of IT equipment and types of
data centers. The number of data center space types has been expanded from the 2007 Report
to account for hyperscale data centers, which are large warehouse-sized data centers that have
emerged with the growth in cloud platforms, mobile devices, social media, and big data.
Previously considered size-based space types are now disaggregated into two categories:
“internal” data centers and “service provider” data centers. Internal data centers represent
traditional facilities that support businesses and institutions while service provider data centers
account for the more specialized facilities that provide the core services of businesses, such as
communication and social media companies. New server classifications are added to
distinguish between servers that only contain one processor socket (“1S”) and those that
contain 2 or more (“2S+”). Additionally, the new server classifications distinguish between
servers that are branded and sold by global computer companies (e.g., Dell, Hewlett-Packard,
IBM), and servers that bypass these brand vendors in the supply chain and are bought directly
from the manufacturing companies that build servers (e.g., Quanta, Wistron, and Foxconn),
often by large Internet companies and typically for use in hyperscale data centers, as shown in
Figure 1. The latter is defined in this report as “unbranded” servers and the former as “branded”
servers. These categories are associated with different usage and data center type placement
assumptions, as described in Section 2.2. The current study also includes updates to previous
storage power consumption estimates, which are now derived from terabyte shipment data and
disaggregated into hard disk drive (HDD) and solid-state drive (SSD) categories. Network
consumption is now estimated through a combination of switch port shipment data of various
speeds and port power consumption from publically available data and published literature. The
equipment types (servers, storage, and networking) modeled in this study are shown in Figure
2.
4
Figure 1. Volume Server Supply Chain
Similar to previous U.S. data center energy estimates,
1
2
3
4
5
this study uses data provided by
the market research firm International Data Corporation (IDC) to derive numbers of data center
servers, as well as storage and network equipment, installed in the United States. Power draw
assumptions are then applied to the estimated installed base of equipment to determine overall
IT equipment energy consumption. IT equipment is disaggregated across various data center
space types, each of which has an associated power use effectiveness (PUE) value. The PUE,
when multiplied by equipment power consumption, gives an estimate of the total power needed
to run the data center, including the data center infrastructure (i.e. cooling, lighting, controls).
Figure 3 shows a schematic of this modeling approach, and Section 2.2 discusses data sources
in more detail.
Figure 2. Equipment Types Modeled in Energy Estimation
\
Unbranded
Vendors
(Foxconn,Quanta,
Wistron,Inventec)
Brand
Vendors
(Dell,IBM,HP)
LargeSpecialized
ServiceCompanies
(Google,Amazon,
Rackspace,Facebook)
Traditi onalCustomers
(universities,banks,
smallcompanies)
hyperscale serversupplychain
5
Figure 3. Schematic of Modeling Approach
2.2 Data
2.2.1 IDC Worldwide Trackers
7
IDC’s Worldwide Quarterly Server, Storage, and Network trackers were used as the basis for
many equipment estimates in this study. Data was obtained in 2014 Q4, and therefore all values
beyond 2014 are considered forecasts. These trackers include historical and forecasted
shipments for each equipment type, as well as installed base estimates for servers.
8
9
10
11
This
data is referred throughout the report as “IDC data”
2.2.2 SPEC Database
15
The SPEC Power benchmark suite, created by the Standard Performance Evaluation
Corporation (SPEC), measures power and performance of servers. SPECpower_ssj2008 is an
industry-standard benchmark application that has been used since 2007, with users self-
submitting results to a database that is reviewed and released to the public quarterly. Data from
2007 to 2015 Q4 was used in this study, and will be referred to as “the SPEC database”.
2.2.3 SERT
12
Database
The Server Efficiency Rating Tool (SERT) was created by SPEC for the U.S. Environmental
Protection Agency’s (EPA) ENERGY STAR program. This tool uses a set of synthetic worklets
to test discrete system components, providing detailed power consumption data at different load
levels. Data from this tool is submitted to the EPA by manufacturers, and is collected and
maintained by the Information Technology Industry Council (ITI). Data collected by ITI through
March 2016 was used in this report, and will be referred to as “the SERT database”.
6
2.3 IT and Infrastructure Equipment Estimates
2.3.1 Server Energy Use
Server Installed Base
Estimates for the U.S. installed base of servers are based on IDC’s “Worldwide Quarterly Server
Tracker – Installed Base”,11 which contains annual data for historical (2006-2014) and
forecasted (2014-2018) server shipments and installed base of servers. These numbers are
provided for each of the three primary server classes (volume, midrange, and high-end) as well
as by processor count (1S or 2S+). The three server classes are an IDC taxonomy based on
average sales value, with volume, midrange, and high-end servers representing < $25,000,
$25,000-$250,000, and >$250,000, respectively. Volume servers typically contain x86
processors and represent the overwhelming majority of servers in data centers. High-end
servers are often large RISC-based systems that operate on Unix and include many high-
performance computing supercomputers. Midrange servers can contain aspects of both volume
and high-end servers. The installed base estimates were extended from 2018 to 2020 using the
compound annual growth rate (CAGR) of the installed base from 2014-2018, the years for which
IDC has forecasted installed base growth.
Each of the 1S and 2S+ categories for volume servers is further disaggregated into “branded”
and “unbranded” classes using additional data provided by IDC and assumptions about future
penetration of hyperscale data centers. First, the percent of all servers located in hyperscale
data centers from 2000-2020 is estimated. (See discussion in Section 2.2.4.) Then, a sigmoid
curve shape (i.e., “S-shape”) is applied to estimate the penetration of unbranded servers within
hyperscale data centers over time. Resulting unbranded installed base estimates for 2000-2020
are calibrated to (1) estimates for 2008-2010, which were provided through communications
with IDC13, and (2) IDC historical server shipment data for 2010-2014, which included
disaggregation by vendor type.8 The assumed hyperscale data center server capacity, sigmoid
curve, data-based estimates, and resulting installed base are shown in Figure 4.
7
Figure 4. Unbranded Server Installed Base and Underlying Assumptions
Figure 5 shows the current and projected total server count for 2000-2020 estimated in this
study, as well as server count estimates from the 2007 Report and from Koomey (2011). The
2007 Report projected much higher server counts than other estimates, with growth from 2007-
2011 following the same trend as 2000-2007, due to these estimates being made prior to the
2008 financial crisis that greatly affected sales. Data from Koomey (2011) is similar to the
current study’s branded server estimates, with server installed base leveling off then slightly
decreasing after 2007. Similarity between these two estimates is expected, as unbranded
servers were not being explicitly tracked by IDC at the time of the Koomey (2011) study. Figure
0
2
4
6
8
10
12
14
16
18
20
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Volumeserverinstalledbase(millions)
Total
Branded
Unbranded
2007Report
IDCdataprovided
forKoomey2011
8
6 shows the volume server installed base estimates from the current study disaggregated by
processor count (1S or 2S+) and vendor type (branded or unbranded).
Server installed base and shipment data provided by IDC was also used to determine the
approximate lifetime of servers. Observation of data showed that for any given year, the number
of servers in the installed base was more than the sum of the previous 4 years’ shipments, but
less than the previous 5. The exact portion of the 5th year’s shipments that were still in the
installed base was found using Equation 1. For 2006-2020, this portion averaged 0.4, with very
little deviation, indicating an approximate server lifetime of 4.4 years. This value was later used
during storage and network installed base calculations based on a report from The Green Grid,
which estimates that servers, storage, and network equipment all have a lifetime of 3-5 years.14
Equation 1



Where f = fraction of 5th year shipments in installed base
IBy = installed base in year y
Sy = shipments in year y
Figure 5. Total Volume Server Installed Base Estimates from Three Studies
0
2
4
6
8
10
12
14
16
18
20
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Volumeserverinstalledbase(millions)
Total
Branded
Unbranded
2007Report
IDCdataprovided
forKoomey2011
9
Figure 6. Volume Server Installed Base 2000-2020
Server Power Draw
Average power use of volume servers was calculated for each year using assumptions for: (1)
maximum power consumption (2) average server utilization and (3) average ability for servers to
scale power use with utilization. The maximum power represents the wattage of the server
when being used at 100% utilization. Average server utilization represents the percent of
computing ability used, on average, over the entire year. Average server scaling ability is the
extent that servers, across the entire installed base, are able to use less-than-maximum power
at non-maximum utilizations.
Maximum power in 2013 for 1S and 2S+ volume servers was estimated from the SERT
database to be 118 W and 365 W respectively.12 These values result in an overall volume
server average maximum wattage of ~330 watts, which is the same assumption used in the
later years of the 2007 Report. Maximum wattage for 2000-2007 was calculated for 1S and 2S+
servers by assuming that the estimates used in the 2007 Report represented a weighted
average of 1S and 2S+ servers and that the ratio between the maximum power of the two types
was the same as the 2013 ratio. Maximum power was held constant from 2007-2020 due to two
observations: (1) the calculated 2007 values for 1S and 2S+ servers were very close to the
SERT (2013) values, and (2) information in the SPEC database15 shows that maximum server
power has remained approximately constant from 2007-2015. While the wattages reported in
the SPEC database were not used directly due to the assumed self-selection bias towards high
efficiency servers in the database, the general temporal trends are assumed to be
representative of all servers. The assumption that the average maximum power remains
constant after 2005 is consistent with assumptions made by Heddgren et al.16 (power draw
based on 50% utilization) and the lower bound assumption by Koomey.3 No difference in
maximum power is assumed between branded and unbranded servers. Power consumption for
0
2
4
6
8
10
12
14
16
18
20
2006 2008 2010 2012 2014 2016 2018 2020
Volumeserverinstalledbase(millions)
Unbranded2+socket
Unbranded1socket
Branded2+socket
Branded1socket
forecast‐‐>
10
mid-range and high-end servers is estimated at the overall average level, with utilization and
scaling assumptions incorporated. These average wattage values for 2000-2007 were taken
from the 2007 Report and extended to 2020 using the 2000-2007 CAGR of approximately 7%,
which is consistent with projections through 2020 provided during industry review. Results are
shown in Figure 7. These wattages assume underlying growth in both utilization levels and
power scaling ability.
Figure 7. Average Power Draw Assumptions for Mid-Range and High-End Servers
Average utilization for volume servers is assumed to vary by space type, as shown in Table 1,
and is assumed to be constant from 2000-2010. A steady increase from 2010 to 2020 is
assumed, to account the prevalence of virtualization opportunities in data centers. Service
provider data centers are assumed to run at higher utilizations than internal data centers, as the
servers in service provider data centers are often configured for more specialized and
predictable operations. Hyperscale data centers are assumed to run at higher utilizations than
other service provider and internal data centers based on server utilization estimates in cloud
and non-cloud data centers.
1
17
18
The values shown in Table 1 represent the average of active
servers, and therefore the inclusion of inactive servers (assumed to be 10% of internal and 5%
of service provider and hyperscale data centers) slightly lowers the overall averages. (See
discussion in Section 4.2 for more information on inactive servers.)
Table 1. Average Active Volume Server Utilization Assumptions
Space Type 2000-2010 2020
Internal 10% 15%
Service Provider 20% 25%
Hyperscale 45% 50%
11
The amount of power consumed at average utilization is dependent on how closely servers
come to achieving power-proportionality, where power consumption scales directly with
utilization. In other words, perfect power-proportionality would mean that a server would only
use 10% of maximum power when run at 10% utilization. One metric used to quantify this
behavior is the dynamic range (DR), which is the ratio between the lowest power level (idle
power) and the maximum power. Minimal documentation exists on the average DR of the U.S.
data center servers stock. The DR varies among different server types and is influenced by
hardware properties, power management software, and settings for server-specific operations,
all of which continue to change. This study estimates the DR of the installed base by first
bounding possible values by a “Maximum DR” trend, which represents the lowest-performing
servers (i.e. those that use a large amount of power at idle) and “Minimum DR” trend, which
represents the highest-performing servers. The assumed Maximum DR is 0.67 in 2007 (the
assumption used in the 2007 Report) and slopes linearly to 0.44 in 2020, as shown in Figure 8.
The SERT database and recent publications
15
18
show servers performing with an average
dynamic range of ~0.44, and therefore this study assumes that, at a minimum, all volume
servers will achieve a DR of 0.44 by 2020. The Minimum DR trend is derived from the SPEC
database values from 2007-2015, which are shown in Figure 9, as this database is generally
understood to represent well-performing servers due to self-selection bias in the server entries.
The SPEC dynamic range trend is modeled as an exponential equation that asymptotically
approaches a DR of 0.1, resulting in the Minimum DR trend shown in Figure 8.
The Maximum and Minimum dynamic range trends provide reasonable bounds of volume server
DR, but an assumption must be made as to where the average DR lies between these bounds.
This study estimates the annual installed base average as a 90/10 mix between the Maximum
and Minimum trends for the Current Trends scenario, resulting in the effective average scaling
trend shown in Figure 8. The DR is assumed to be constant across vendor types (branded and
unbranded) and processor counts (1S and 2S+).
12
Figure 8. Assumed Dynamic Range of Volume Servers
Figure 9. Dynamic Range of 1- and 2-Socket Servers in SPEC Database
Dynamic range and maximum wattage values are used to calculate the slope of the utilization
versus power consumption curve for volume servers in each year, as shown in Equation 2. This
slope is then used with utilization assumptions to calculate the average wattage of volume
servers in each space type in each year. This calculation is shown in Equation 3, and results are
shown in Figure 10. These average wattage values, along with installed base estimates, provide
the total annual server energy consumption estimates shown in
Figure 11.
Equation 2
P
∗ 1 
Where m = slope of utilization vs. power line
P
max
= maximum server wattage
DR = dynamic range of server (fraction of max power used at idle)
Equation 3
 
 ∗1
Where P
avg
= average server wattage
m = slope of utilization vs. power line
u
avg
= average server utilization
13
Figure 10. Maximum and Effective Average Power Estimates for Volume Servers
The solid black line represents maximum power of each server size. Servers in hyperscale data centers
use more power than those in internal or service provider due to higher utilizations.
Figure 11. Total U.S. Annual Direct Server Electricity Consumption by Server Class
14
2.3.2 Storage Energy Use
The installed base of data storage equipment was calculated in terabyte (TB) units using IDC
historical (2010-2014) and forecast (2015-2019) shipment data. This data includes external
storage units from both branded and unbranded vendors as well as internal storage on servers
with more than two storage drives. This study estimates the energy consumed by these
supplemental storage types collectively, and refers to all types included in the IDC tracking data
as either “Data Center Storage” or “External Storage”. The provided shipment data is extended
back to 2000 and forward to 2020 using the compound annual growth rate (CAGR) from 2010-
2019. Then, installed base is calculated assuming an average storage life of 4.4 years, equal to
the estimated lifetime of servers (see Section 2.2.1).
14
This installed base is then disaggregated
into hard disk drive (HDD) and solid-state drive (SSD) storage categories using a 2015
ASHRAE report that estimated that in SSD accounted for 8% of non-tape storage in 2012, and
would grow to 22% by 2017.
20
The final installed base estimate in TB is shown in Figure 12.
Figure 12. Total U.S. Data Center Storage Installed Base in Capacity (TB)
Power consumption of HDDs is relatively fixed on a per-disk level, and is generally not
dependent on the capacity of the disk.
3 19
For SSD units, power consumption is more closely
related to capacity (more specifically, to read/write frequency), but is still often reported on a
per-disk level. Therefore, it is useful to convert installed base estimates of both HDD and SSD
storage from capacity (TB) to number of drive units. To do so, the trend of TB/drive over time for
each storage type is first estimated. For 2000-2007, the TB/drive metric for HDD is calculated
from (1) this study’s installed base estimate in TB, for external branded drives only, and (2) the
2007 Report’s disk installed base data, which was provided by storage manufacturers at the
time. In 2020, HDDs are assumed to provide an average capacity of 10 TB/disk from drive
capacity improvements and capacity optimization methods, based on industry feedback. Data
between the 2007 and 2020 estimates are populated using an exponential trend, resulting in the
trend shown in Figure 13. For SSD, an average capacity of 5 TB/drive is assumed to be
15
reached by 2020 based on industry feedback, and the same annual growth rate seen in HDD
TB/disk (approximately 27%) is assumed for SSD for all years prior, as shown in Figure 13.
Annual TB/drive estimates are multiplied by the TB installed base estimate (Figure 12) to get the
estimated storage installed base in number of drives. This estimate is shown in Figure 14, which
shows growth in number of installed drives through 2018, after which point HDD count begins to
decrease (SSD count continues to grow). Note that this is the number of installed hard disk
drives, and does not indicate slowing storage demand. Decreasing drive count is due to HDD
capacity per disk (TB/disk, Figure 13) growing at a faster rate than projected HDD capacity
demand (TB, Figure 12). SSDs make up 47% of the installed drive base in 2020, which is within
the 40-50% range estimated by industry.
Figure 13. Estimated Average Capacity of U.S. Data Center Storage Drives
Figure 14. Total U.S. Data Center Storage Installed Base in Drive Count
16
Power consumption for HDDs was assumed to be 14 W/disk from 2000-2006 as estimated by
Seagate for the 2007 Report. A 2015 ASHRAE report
20
stated minimum, average, and
maximum power usage for various speeds and sizes of HDDs, while a NYSERDA report
21
estimated the breakdown of these various drive types in the installed base. Using these two
reports, the average wattage of HDDs is estimated to be 8.6 W/disk in 2015. This represents a
5% annual decrease in disk wattage from 2006-2015, which is assumed to continue through
2020, resulting in 6.5 W/disk as shown in Figure 15. This improvement in disk efficiency
represents more efficient disk drive components, lower power use in idle states, and use of
capacity optimization methods.
SSD wattage is assumed to be constant at 6 W/drive, as shown in Figure 15, based on both
IDC
22
and ASHRAE
20
reports. Constant drive wattage paired with our TB/drive estimates aligns
with industry expectations that SSD capacity per watt will increase three to four-fold by 2020.
Figure 15. Average Wattage of Storage Drives in U.S. Data Centers
Electricity consumption for all data center storage is calculated as the product of the estimated
installed base of drives and the assumed power consumption per drive. Additional operational
consumption was then added to account for the controller and associated components required
to operate external storage systems. This operational energy is assumed to equal 25% of the
storage energy, based on industry comment, and only applies to storage that is external to
servers. Resulting total storage energy consumption is calculated according to Equation 4 and
results are shown in Figure 16.
17
Equation 4

, ∗
,
, ∗1

Where E
y
= Storage electricity consumption in year y
I
t,y
= Installed base of storage type t in year y
P
t,y
= Per-unit power consumption of storage type t in year y
h
y
= Number of hours in year y
O = Operational energy as a fraction of storage energy
C
external
= Capacity of the external storage installed base
C
total
= Capacity of the total storage installed base
Figure 16. Total U.S. Data Center Storage Electricity Consumption
2.3.3 Network Energy Use
Previous reports estimated network energy use either as a percentage of server energy use
1
16
or as a product of number of servers, ports per server, and watts per port.
2
This study expands
upon these previous efforts by using port installed base estimates disaggregated by port speed
along with port wattage estimates for each speed, as described below.
Network equipment data from IDC includes historical (2008-2014) and forecasted (2015-2019)
shipments of network ports disaggregated into four port speed categories: 100 MB, 1000 MB,
10 GB, and 40 GB. Similar to storage installed base calculations, this shipment data is
extrapolated to estimate number of shipments for 2000-2007 and 2020, then used to calculate
the installed base of ports by assuming an average network equipment life of 4.4 years, as
estimated from server data (see Section 2.2.1).
14
Resulting port installed base estimates are
shown in Figure 17.
18
Figure 17. Total U.S. Data Center Installed Base of Network Ports
The power draw of network equipment was estimated on a per-port basis using different values
for each port speed. Per-port wattage trends were estimated based on: the 2007 Report, which
estimated an average of 8 watts (W) across the installed base; a 2012 empirical study on
network energy in small to medium sized data centers,
23
which reported estimates of 1.4, 2.3,
and 3.6 W for 100 MB, 1000 MB, and 10 GB ports, respectively; a survey conducted as part of
this study of 51 technical specification sheets from network manufacturers, which indicated that
40 GB ports use approximately 1.7 times the power of 10 GB ports; and industry comment,
which suggested 1 W/port for 1000 MB ports in 2020. These values were used to create the
estimated trends shown in Figure 18. The approximate range of W/port values for various port
speeds applied in this study is consistent with power estimates in previous network studies.
24
25
Figure 18. Assumed Network Power for Four Port Speeds
19
Total estimates of network energy consumption were calculated as the product of the number of
ports in the installed base and the assumed per-port power draw for each port speed, as shown
in Equation 5. Results are shown in Figure 19. While other types of network equipment besides
Level 2/Level 3 network ports exist in data centers, detailed shipment and power consumption
data are not available for these devices, and this equipment is assumed to contribute minimally
to overall data center power consumption.
Equation 5

, ∗
, ∗
∈
Where E
y
= Network electricity consumption in year y
S = set of port speeds: 100 MB, 1000 MB, 10 GB, 40 GB
N
s,y
= Number of installed ports of speed s in year y
P
s,y
= Power consumption of ports of speed s in year y
h
y
= number of hours in year y
Figure 19. Total U.S. Data Center Network Equipment Electricity Consumption
2.3.4 Data Center Space Type Classifications
Installed base estimates of servers, storage, and network equipment were disaggregated into
11 different data center building space types: five internal data center types that include server
closets, server rooms, localized data centers, mid-tier data centers, and high-end data centers,
as well as six service provider data center types that include server closets, server rooms,
localized data centers, mid-tier data centers, high-end data centers, and hyperscale data
centers. Internal data centers represent facilities operated by an organization for internal
20
activities, including email and productivity software, and are typically associated with finance,
education, and other institutions not directly involved with providing IT services. Large internal
data centers are often referred to as enterprise or corporate data centers. Service provider data
centers (also called production data centers, and inclusive of colocation facilities) contain IT
equipment used to provide communication services often associated with the core product of a
business, such as the services provided by Google, Amazon, Facebook, and other
telecommunication and social media companies. Service provider data centers include the
same five size categories as internal data centers, as well hyperscale sized data centers, which
are sometimes referred to as “mega” or “warehouse scale” data centers and are often
associated with cloud data centers. These types of spaces are defined by IDC as shown in
Table 2.
Space disaggregation is necessary to estimate energy use, as half of U.S. servers are located
in server closets and server rooms26 that can have significantly different IT equipment and
infrastructure characteristics than larger data centers, and because service provider data
centers associated with cloud platforms, mobile devices, social media, and big data have grown
significantly in recent years.27 In the context of this study, “infrastructure” consists of the data
center equipment that is not used solely for the purpose of performing computations or the
storage or transmission of data. This includes cooling systems, lighting, power supplies,
security, and so on. Determining the distribution of different servers across space types allows
for more accurate characterization of the total energy use associated with different server
environments. It also allows for better characterization of electricity costs because most server
closets, server rooms, and localized data centers are expected to be subject to commercial
electricity rates, whereas larger mid-tier and enterprise-class data centers are expected to be
subject to industrial electricity rates. Disaggregation into the eleven space types is a significant
expansion on the 2007 study’s five space types. An overview of the methods used to allocate IT
equipment to the various data center types are shown in Table 3, with detailed descriptions
provided below.
The number of data centers in each space category in the U.S. for 2005 was derived from a
2006 IDC report,26 while 2012-2018 estimates were taken directly from a 2014 IDC report.27 The
2005 numbers only included five space categories, and therefore were split into internal and
service provider categories (based on the 2012 ratios). No hyperscale data centers were
assumed to exist in 2005. These 2005 numbers were then assumed to grow linearly to 2012
values. The number of data centers for 2000-2004 and 2019-2020 were estimated using the
2005-2009 (as reported by Bailey26) and 2012-2018 CAGRs, respectively, for each of the eleven
room types.
Bailey26 also reported average number of servers per data center for five space categories in
2005. Servers per data center for each of the five space sizes were assumed to be applicable to
both internal and service provider data centers in that category. These numbers were then
adjusted for each year so that, when multiplied by the number of data centers, the total number
of servers was consistent with this study’s estimates of total server installed base (excluding
servers assigned to hyperscale data centers). The resulting total number of servers in each
space category are shown in Figure 20.
21
Table 2. Typical IT Equipment and Site Infrastructure System Characteristics by Space Type
Space type Typical size Typical infrastructure system characteristics
Internal server closet < 100 ft2 Often outside of central IT control (often at a
remote location) that has little to no dedicated
cooling.
Internal server room 100-999 ft2 Usually under IT control, may have some
dedicated power and cooling capabilities.
Localized internal
datacenter
500-1,999 ft2 Has some power and cooling redundancy to
ensure constant temperature and humidity
settings.
Midtier internal
datacenter
2,000-19,999 ft2 Superior cooling systems that are probably
redundant.
High-end internal
datacenter
> 20,000 ft2 Has advanced cooling systems and redundant
power.
Point-of-presence
server closet
< 100 ft2 At local points of presence for OSS and BSS
services. Typically leverages POP power and
cooling. Space is often a premium.
Point-of-presence
server room
100-999 ft2 Secondary computer point of presence for OSS
and BSS services. Typically leverages POP power
and cooling.
Localized service
provider datacenter
Including
subsegment:
containerized
datacenter
500-1,999 ft2 Has some power or cooling redundancy to ensure
constant temperature and humidity settings. These
are typically facilities set up by VARs to provide
managed services for clients.
Midtier service
provider datacenter
Including
subsegment:
prefabricated
datacenter
2,000-19,999 ft2 Location for small or midsize collocation/hosting
provider. Also includes regional facilities for
multinational communications service providers.
Has superior cooling systems that are probably
redundant.
High-end service
provider datacenter
> 20,000 ft2 Primary server location for a service provider. May
be subdivided into modules for greater flexibility in
expansion/refresh. Has advanced cooling systems
and redundant power.
Hyperscale
datacenter
Up to over
400,000 ft2
Primary server location for large collocation and
cloud service providers. Based on modular
designs, with individual modules of 50,000 sq ft on
average in up to 8 modules. Employs advanced
cooling systems and redundant power.
While IDC shipment estimates include servers destined for hyperscale data centers, the
percentage of servers that are located in hyperscale data centers was not available. Therefore,
this study uses a simple estimation that the percent of servers housed in hyperscale data
centers will grow linearly from 0% in 2005 to 40% in 2020. This assumption is a conservative
22
estimate of future unbranded server installed base as calculated by shipment forecasts, and
aligns with other industry projections of the unbranded server market28 and expectations that at
least 40% of the data in 2020 is expected to be “touched” by the cloud at least once.29 This
estimate also aligns with current IDC estimates of the global average of servers per data center
for high-end and hyperscale data centers.30
Table 3. Allocation of Data Center Equipment Across Space Types
Step Equipment Allocation Method
1 Total Servers
Set percentage (varies annually) to Hyperscale
Remaining based on estimated data center counts
and 2005 servers per data center estimate
2 Midrange Servers
5% Server Rooms
30% Localized and Mid-tier Data Centers
65% Enterprise Data Centers
3 High-End Servers
30% Localized and Mid-tier Data Centers
70% Enterprise Data Centers
4 Unbranded 1S and 2S+
Volume Servers
100% Hyperscale Data Centers
5 Branded 2S+ Volume
Servers
Fill remaining spots in Hyperscale
6 Branded 1S and 2S+
Volume Servers
Fill remaining spots in all other data centers, keeping
1S and 2S+ in equal proportion
7 Storage
None in Server Closets or Rooms
Allocated to all other spaces based on total server
count
8 Network Ports
Total allocated based on total server count, with
higher speeds trending towards larger data centers
23
Figure 20. Total Server Installed Base by Data Center Space Category
Once total server installed base is allocated to various room types, the prevalence of each of
the six server classifications in this study (volume branded 1S and 2S+, volume unbranded 1S
and 2S+, mid-range, and high-end) is allocated to the various space categories. Mid-range and
high-end servers were allocated based on a set of assumptions used in the 2007 Report: (1) no
mid-range or high-end servers are in server closets (2) 5% of mid-range are in server rooms
and (3) 65% and 75% of mid-range and high-end servers, respectively, are in high-end data
centers. The remainder of mid-range and high-end servers were split evenly between localized
and mid-tier data centers. Within a space type, servers were split between internal and service
provider space categories based on the total number of data centers in each category. All
volume unbranded servers were assumed to be in hyperscale data centers. Lastly, branded
volume servers were assumed as the difference between the total number of servers assigned
to a given space category, and the number of mid-range, high-end, and unbranded servers
assigned in that category.
External storage is assumed to be present in all data center space types except server closets
and server rooms, and is allocated based on the number of servers in each space type in the
given year. Network ports are also distributed among space types in direct proportion to the
number of servers in the given space type. While total number of ports per server is constant
across space categories (in a given year), the speed of the ports installed varies between space
categories. Space categories were grouped into “fast” (hyperscale and high-end), “medium”
(mid-tier and localized), and “slow” (rooms and closets), and the installed base of ports was
distributed accordingly.
Infrastructure energy consumption (cooling equipment, uninterrupted power supplies, lighting,
etc.) is calculated using the power usage effectiveness (PUE) metric.
31
This metric represents
the total energy required by the data center in relation to the energy needed for the IT
equipment. A data center with PUE of 1 would use no electricity other than the IT equipment. At
24
the time of the 2007 Report, the average PUE was estimated to be 2.0, indicating that non-IT
energy use was about equal to the IT energy use in a data center. Other studies at that time
showed many data centers with PUE values greater than 3.0.32 33 34 More recent studies show
that while a wide range of PUE values is still observed, the average PUE has only modestly
improved to about 1.8-1.9.35 36 The slower rate of efficiency improvement in PUE relative to IT
equipment is partially due to the slower turnover rate of a data center’s infrastructure relative to
the IT equipment. The opportunities to improve data center PUE increase with larger data
centers that have the ability to develop better airflow management and employ more efficient
cooling equipment or advanced cooling technologies such as liquid cooling. Consequently,
smaller data centers are still being measured with PUE values greater than 2.037 while large
hyperscale cloud data centers are beginning to record PUE value of 1.1 or less.38 39 40 Table 4
presents the characteristic infrastructure equipment and corresponding average PUE values
assumed for each data center size for 2014. PUE values for each data center size are based on
previously published values, expert elicitation, and energy modeling results for different data
center infrastructure configurations.41 For the current trends scenario, average PUE values for
each size data center are assumed to be the same for internal and service provider data centers
and, given the very modest improvements observed in published data for average PUE values,
to improve by 1% per year through 2020, except for closets. The PUE for closets is held
constant at 2.0 throughout the analysis period, given that the lack of dedicated cooling and
electrical equipment in this space type limits opportunity for improvement. Applying the PUE
values in Table 4 to IT energy use in each data center type provides the total energy
consumption by data center component (Figure 21) and by space type (Figure 22).
Table 4. 2014 PUE by Space Type
Space Type IT Transformer UPS Cooling Lighting Total PUE
Closet 1 0.05 - 0.93 0.02 2.0
Room 1 0.05 0.2 1.23 0.02 2.5
Localized 1 0.05 0.2 0.73 0.02 2.0
Midtier 1 0.05 0.2 0.63 0.02 1.9
High-end 1 0.03 0.1 0.55 0.02 1.7
Hyperscale 1 0.02 - 0.16 0.02 1.2
2.3.5 Total Data Center Energy Consumption
As shown in Figure 21, electricity consumption of data centers has been relatively flat in recent
years, which is attributable to many factors. The growth rate of server shipments has diminished
over the past 15 years. From 2000-2005, server shipments increased by 15% each year
resulting in a near doubling of servers operating in data centers. From 2005-2010, the annual
shipment increase fell to 5%, most likely driven by the economic recession as well as the
emergence of server virtualization during that period. The annual growth in server shipments
further dropped after 2010 to 3% and that growth rate is now expected to continue through
2020. This 3% annual growth rate coincides with the rise in hyperscale data centers and an
increased popularity of moving previously localized data center activity to colocation or cloud
25
facilities. In fact, nearly all server shipment growth since 2010 occurred in servers destined for
hyperscale data centers, where servers are often configured for maximum productivity and
operated at higher utilization rates, resulting in fewer servers needed than would be required to
provide the same services in traditional, smaller, data centers.
Along with total server count, the power demand for each server has also changed. While
server power requirements were observed to be increasing from 2000-2005, power demand
appears to have stayed fairly constant since 2005. Additionally, servers are improving in their
power scaling abilities, thus reducing power draw during idle periods or when at low utilization.
Efficiency improvements in storage, network and infrastructure also influence the electricity
estimates in this report. Storage devices are becoming more efficient on a per-drive basis, with
the growth in drive storage capacity projected to outpace increases in data storage demand by
2020, ultimately reducing the number of physical drives needed throughout data centers.
Recent estimates of network port power consumption are now much lower than estimates from
the past decade. Increased awareness in data center infrastructure operations (e.g. cooling) has
resulted in improved efficiency across data center types, though the most significantly in large
cloud data centers that are innovatively designed to maximum infrastructure efficiency.
Figure 21. Total Electricity Consumption by Technology Type
26
Figure 22. Total Electricity Consumption by Space Type
Results from this study for 2000-2014 are shown in Figure 23 alongside results from three
previous studies: the 2007 Report; Masanet et al. (2011), which built upon the 2007 Report to
estimate consumption in 2008; and Koomey (2011), which estimated 2010 consumption using
updated server shipment data after the 2009 financial crisis. While data center operations since
2008 likely benefited from some adoption of energy efficiency measures, such as increased
server consolidation and improved IT power management, the distinct reduction in electricity
use observed in the later studies immediately after 2008 was likely also due to a lower installed
server base associated with the economic recession and the related efficiency improvements
driven by pressure to cut energy and equipment costs.
Figure 23. Historical Data Center Total Electricity Use
27
Growth in data center energy consumption has slowed drastically since the previous decade.
However, demand for computations and the amount of productivity performed by data centers
continues to rise at substantial rates.42 Technology advancements have made IT equipment
more efficient by being able to perform more work on a given device, while other design and
management efforts have made the industry more energy efficient. A counterfactual scenario
was created for this study that estimates what data center energy consumption would have
been if industry energy-savings efforts were halted in 2010. The metrics frozen in 2010 include:
Industry-wide average server utilization remains at 14%
Dynamic range of server remains at 0.59 (i.e., servers use 59% of their maximum power
at idle) (see Figure 8)
Wattage per HDD remains at about 11.3 watts (see Figure 15)
Wattage of network ports remains at 1.6, 2.6, 4.1, and 7.1 for 100Mb, 1000Mb, 10Gb,
and 40 Gb, respectively (see Figure 18)
Industry-wide weighted average PUE remains at 1.89
This scenario does not halt the technological advancements of the computing industry in terms
computational performance (i.e., computations/second per servers) and the electrical efficiency
of computations (i.e., computations per kWh). Computational performance and the electrical
efficiency of computations are assumed to continue to improve in parallel and at similar rates43
(thereby cancelling each other out), since the main trend driving both advancements are smaller
transistors.42 Additionally, wattage per HDD and per network port are held at 2010 levels but
storage capacity (i.e., TB per drive) and a shift towards fasters ports are assumed to progress
as normal.
The results of this scenario are shown in
Figure 24, with 2014 energy use that is 60% higher than currently estimated, and projected
2020 use 170% higher than the Current Trends scenario. Energy savings of the industry are
therefore estimated to be 100 billion kWh from 2010-2014, and an additional 520 billion kWh
from 2015-2020. The overwhelming majority of these savings come from the servers and
infrastructure. Server savings are driven by the increase in industry-wide average utilization that
results from consolidation efforts and the growth data centers that operate at higher utilization
levels (i.e., hyperscale and other service provider data centers). Infrastructure savings result
from the reduced amount of IT equipment that require cooling and electrical services as well as
the decrease in industry-wide average PUE, brought down by the growth in data centers with
very low PUE values (i.e., hyperscale data centers).
28
Figure 24. Data Center Electricity Consumption in Current Trends
and 2010 Energy Efficiency Scenarios
The 2010 Energy Efficiency scenario assumes that data center energy-related design and operational
efforts do not continue past 2010, which indicates that current trend energy efficiency practices will have
saved 620 billion kWh of electricity over the period 2010-2020.
2.3.6 Data Center Water Consumption
Along with electricity demand, data centers require significant water consumption during
operation. Water is consumed at two key points during data center operation.
44
First, water is
required during the generation of electricity from primary energy that is eventually transmitted
for use at the data center site. A national average of 7.6 liters (2.0 gallons) of water are
consumed for each kWh when weighting the water losses at both thermoelectric and
hydroelectric plants in the U.S.
45
Second, in medium to larger size data centers that employ
cooling tower based chillers to improve energy efficiency, water is consumed at the data center
site itself. Cooling towers use water evaporation to reject heat from the data center causing
losses approximately equal to the latent heat of vaporization for water, along with some
additional losses for drift and blowdown. In larger data centers this on site water consumption
can be significant, with data centers that have 15 MW of IT capacity consuming between 80-130
million gallons annually.
46
47
In this study, on-site water consumption is estimated at 1.8 liters
(0.46 gallons) per kWh of total data center site energy use for all data centers except for closet
and room data centers, which are assumed to use direct expansion (air-cooled chillers). With
these assumptions, approximately 626 billion liters of water was estimated to be consumed in
2014 for data centers, with that number reaching 660 billion liters in 2020. Data center water
consumption is shown in Figure 25 and Figure 26. Note that water consumption associated with
the generation of electricity varies significantly by primary energy type and power plant
29
efficiency, such that the actual water consumption attributable to any individual data center will
depend on its specific location and electricity provider.
Figure 25. Direct vs. Indirect U.S. Data Center Water Consumption
Figure 26. Total U.S. Data Center Water Consumption by Space Type
3 Energy Use Associated with Federal Government Servers
and Data Centers
The Federal Data Center Consolidation Initiative (FDCCI), established in 2010 to reverse the
historic growth of Federal data centers, has been conducting surveys on federally operated data
centers to verify existing records and expand inventory data for these facilities (e.g. address,
operational status and cost, square footage, server count, etc.).
48
For this study, server count
30
estimates were provided by the Office of Management and Budget (OMB) for 2012-2014,
corresponding to approximately 1-2% of nationwide server installed base estimates, as shown
in Table 5.
Table 5. Servers in Federal Data centers Tracked by OMB
2012 2013 2014
Total Server Count 238,709 295,316 118,851
Percent of National Installed Base 1.8% 2.1% 0.8%
Estimates provided in Table 5 do not capture shipments of servers to facilities that were
managed by contractors for the federal government and adequate information to estimate the
percent of federal server use that occurs in these facilities is not available. Therefore, these
numbers provide only a portion and not a complete representation of the impact of the federal
government associated with data center operation.
This estimate from OMB is a sharp decrease from the 2007 Report where federal servers and
data centers were estimated to be about 10% of the U.S. total. At that time, no data could be
found in the public domain on the number of servers or data centers operated by (or for) the
federal government and therefore the server estimate was based on interviews conducted with
major U.S. manufacturers. These interviews led to the conclusion that server shipments to the
federal government represented about 5-10% of annual U.S. server shipments and the higher
end of this range was used in an attempt to account for servers shipped to government
contractors that were not counted as federal shipments.
FDCCI efforts have resulted in the closing of approximately 1.7 million square feet of data
center space since 2010.49 According to current estimates of server density in various space
types this correlates to the removal of approximately 60,000 servers from the federal inventory
or about half of the 2014 estimate. However, even without these consolidation efforts the 2014
server count would correspond to only 1.2% of the national installed base. This large
discrepancy compared to the 2007 Report estimate implies OMB’s server accounting may not
be capturing a significant portion of federal servers in the U.S. data center stock.
No data on federal server shipments by server class (i.e., volume, mid-range, and high-end) are
available. It is possible that the federal government accounts for a significant fraction of high-
end server electricity use, given that about one third of the 100 largest supercomputers in the
world are housed in U.S. government-owned facilities, including four out of the top ten in the
world.50 This larger representation of high-end servers would potentially put federal energy
consumption at a higher percentage of the national installed base than the percentage
represented in Table 5.
31
4 Expected Energy Savings Opportunities
4.1 Energy Efficiency Trends
A number of energy efficiency trends have been shaping server operations over the past
decade, including virtualization and consolidation, which are expected to continue to generate
energy savings and reduce server footprint. Physical parts of the server such as the
microprocessor, cooling fan, and power supply are also improving in energy efficiency (or being
eliminated, such as the fan in servers with liquid cooling), thus further reducing server power
consumption per unit of computing output. More efficient storage devices such as SSDs are
coming down in cost and their increasing prevalence will continue to drive efficiency.
Furthermore, the growing trend of cloud computing has resulted in significantly larger data
centers that are more efficient, both in terms of server utilization and infrastructure PUE,
compared to traditional enterprise data centers. Along with the current trend previously
described in this report, three additional scenarios and two combinations of scenarios were
modeled to estimate near-term energy efficiency opportunities in U.S. data centers.
4.2 Improved Management Scenario
The Improved Management (IM) scenario consists of two components: improved PUE and
removal of inactive servers. Improved PUE represents an effort for smaller data centers to
reduce their infrastructure (e.g. cooling, lighting) energy demand through improved airflow and
thermal management, such hot/cold isle isolation and reducing set point temperatures. PUE
values trend linearly after 2014 to the improved numbers shown in Table 6. Inactive servers
(also referred to as comatose or “zombie” servers) represent obsolete or unused servers that
consume electricity but provide no useful information services. Previous studies have estimated
that inactive servers represent 10-30% of servers in U.S. data centers.51 52 53 54 Inactive server
removal represents an impact of raised awareness on part of data center operators as to what
equipment is being utilized in the data center. In order to model this impact, servers are first
assigned to each space type as described in Section 2.2.4, then the number of volume servers
in internal data centers is decreased by 10%, and in service provider data centers by 5%, for
each year after 2014. The percent decrease in service provider data centers is assumed to be
smaller because these data centers tend to have a lower rate of inactive servers due to better
management practices that avoid the institutional problems of dispersed responsibility between
IT and facility departments which often plagues internal data centers.55
The IM scenario results in total energy savings in 2020 relative to the Current Trends scenario
of 10% which is composed of: 2.5% from server energy savings due to the removal of inactive
servers; no savings in storage or network energy; and 7.5% from savings in infrastructure
energy due to the reduced server energy along with improved PUE.
32
Table 6. PUE and Redundancy Values for Efficiency Scenarios
Space Type 2014 PUE
2020 PUE
Redundancy
Current
Trends
Improved
Management
Best
Practices
Closet 2.0 2.00 2.00 2.00 N+0.5N
Room 2.5 2.35 1.70 1.50 N+1
Localized 2.0 1.88 1.70 1.50 N+1
Midtier 1.9 1.79 1.70 1.40 N+0.2N
High-end 1.7 1.60 1.51 1.30 N+0.5N
Hyperscale 1.2 1.13 1.13 1.10 N
4.3 Best Practices Scenario
The Best Practices (BP) scenario builds upon the improvements in the IM scenario. These best
practices include further improved PUE values (Table 6) from using more efficient infrastructure
components and employing economizer or liquid cooling when applicable, as well as improved
dynamic range (power scaling ability) of servers, server and network consolidation efforts, and
reduced storage disk and network port power consumption.
In all scenarios, average dynamic range is calculated by determining what mix of servers follow
a minimum versus a maximum dynamic range trend (Figure 8). For the CT scenario, the
average dynamic range of volume servers is assumed to be represented by 90/10 mix of the
maximum and minimum dynamic range trends. The BP scenario has a more aggressive
penetration of servers that follow the maximum dynamic range trend and assumes that this ratio
reaches 50/50 by 2020. This results in the dynamic range trend shown in Figure 27, where
volume servers reach a dynamic range of 0.28 in 2020. While these number are possible, it
should be noted that getting below 20-25% of maximum power at idle can require powering
down specific functionality that may increase idle-to-active wake-up latency and result in
delayed response times. Attempts to improve the idle power savings versus latency include
establishing different levels of inactive modes with varying degrees of power savings and
response times.56 57 As discussed in Section 2.2.1, these numbers are used to calculate the
slope of the utilization versus power line (Equation 2) and therefore the power consumption at
average utilization levels (Equation 3).
Server consolidation entails replacing multiple servers running at low utilization with a single
server running at a higher utilization. Eighty percent of the volume server installed base is
assumed to be consolidated onto servers that operate at the utilization levels presented in Table
7. These parameters are chosen based on the assumption that servers in internal and service
provider data centers could be consolidated through methods such as virtualization and
containerization to the current high end of utilization observed in servers in small and medium
size data centers17 18 and servers in hyperscale data centers could be consolidated to 75%
utilization through improved CPU bandwidth provision techniques, which have been shown to
achieve average data center server utilizations as high as 90%.58 There is a small additional
benefit to consolidation by reduction in redundant servers required, which is calculated
33
according to the data center size-specific redundancy frameworks presented in Masanet
(2013),
59
shown in Table 6. The formulas in the “Redundancy” column represent the total
number of servers needed for a data center containing N functional servers. For example,
redundancy of “N+1” means that there is one redundant server present in each data center,
while redundancy of “N+0.1N” means that there is one redundant server for every 10 functional
servers. For data centers where the number of redundant servers scales with server count (i.e.
closets, mid-tier, and high-end enterprise), consolidation of servers reduces the number of
redundant servers required.
Figure 27. Average Volume Server Dynamic Range for
Current Trends and Best Practices Scenarios
Table 7. Best Practices Scenario Consolidation Parameters
Space Type
Percent of installed
base consolidated
by 2020
Utilization Consolidation
utilization
overhead
Pre-
consolidation
Post-
consolidation
Internal 80% 10-15% 45% 5%
Service Provider 80% 20-25% 55% 5%
Hyperscale 80% 45-50% 75% 5%
Equation 6 describes the number of servers that are consolidated each year from 2015-2020.
Equation 7 describes the number of pre-consolidation servers that can be replaced by each
post-consolidated server, based on pre- and post- consolidation utilization, utilization
“overhead”, and redundancy considerations. Utilization overhead is estimated as 5% per post-
consolidation server and accounts for the applications that must be run on the server to balance
multiple workloads. In essence, this means that consolidation of two servers running at 10%
utilization would result in one server running at 25% utilization. Because the scenario’s post-
34
consolidation utilizations are defined, Equation 7 is used to calculate the number of pre-
consolidation servers that are replaced by each post-consolidation server. The number of
servers consolidated each year (Equation 6) is then divided by this metric to determine how
many servers (out of those that are consolidated) remain in the installed base, as shown in
Equation 8. Finally, the new post-consolidation installed base is calculated as shown in
Equation 9. Consolidation impacts on server installed base are shown in Figure 28.
Equation 6
, 
∗

∗
Where Scon,y = number of servers to be consolidated in year y
Cyf = Percent of installed base consolidated in final year
yi = year consolidation begins
yf = year consolidation ends
IBy = baseline installed base in year y
Equation 7
  
 1

Where Ncon = number of servers that can be consolidated into one
upost = post-consolidation server utilization
uo = consolidation utilization overhead
upre = pre-consolidation server utilization
r = fraction of pre-consolidation servers that are non-redundant
Equation 8
, ,

Where Sres,y = number of consolidated servers resulting in year y
Scon,y = number of servers consolidated in year y
Ncon = number of servers that can be consolidated into one
Equation 9
, 
, ,
Where IBy,post = installed base in year y post-consolidation
IBy = baseline installed base in year y
Scon,y = number of servers to be consolidated in year y
Sres,y = number of consolidated servers resulting in year y
35
Figure 28. Volume Server Installed Base for Current Trends and Best Practices Scenarios
Network port consolidation is similar to server consolidation. Eighty percent of the installed base
of 10 GB ports are assumed to be consolidated into 40 GB ports, which can transmit 4 times the
data at only 1.7 times the power. In addition, the BP scenario assumes that ports installed in
high-end enterprise and hyperscale data centers will become 25%
60
more efficient by 2020. This
reduction in port wattage is based on network improvements in network topology, dynamic link
rate adaptation, and link and switch sleep modes.
25
BP scenario impacts on port installed base
are shown in
Figure 29.
Figure 29. Network Installed Base for 10 GB and 40 GB Ports in
Current Trends and Best Practices Scenarios
36
The BP scenario also includes a provision to improve the efficiency of storage disks by 25% in
2020. This improvement increases linearly, and is based upon a scenario where future storage
disks have reduced energy usage in their idle state. The resulting disk power consumption is
shown in Figure 30.
Figure 30. Storage Disk Power Consumption for Current Trends and Best Practices Scenarios
The BP scenario results in total energy savings in 2020 relative to the Current Trends scenario
of 40% which is composed of: 15% from server energy savings due to the removal of inactive
servers, the improvement in server scaling, and consolidation of servers; 3% from savings in
storage energy due to improved disk efficiency, 1% from savings in network energy due to port
consolidation; and 22% from savings in infrastructure energy due to the reduced IT energy
along with improved PUE.
4.4 Hyperscale Shift Scenario
The Hyperscale Shift (HS) scenario involves the consolidation of 80% of the servers in non-
hyperscale data centers into hyperscale data centers, excluding servers in Service Provider
Rooms and Closets, which are defined to include local point-of-presence facilities that would still
be needed to support management functions for hyperscale data centers.
27
The 80% shift is in
addition to the shift already accounted for in the Current Trends scenario. Hyperscale data
centers operate servers at higher utilizations in infrastructure-efficient (low-PUE) spaces, and
include cloud-based platforms that remove the need for dedicated redundancy servers, so that
consolidating IT services from many small disparate data centers into hyperscale data center
can yield significant energy savings. HS scenario impacts on volume server installed base are
shown in Figure 31. As shown, the installed base in non-hyperscale data centers decreases to
approximately one quarter of the CT scenario value by 2020. The impact on the installed base
in hyperscale data centers is smaller since, on average, one server in a hyperscale data center
can replace 3.75 servers in non-hyperscale data centers. This is because servers in hyperscale
37
data centers are assumed to run at roughly 3 times the utilization of non-hyperscale data
centers and have no redundancy requirements.
61
Figure 31. Volume Server Installed Base in Current Trends and Hyperscale Shift Scenarios
The HS scenario results in total energy savings in 2020 relative to the Current Trends scenario
of 25% which is composed of: 10% from server energy savings, due to the consolidation; no
savings in storage or network energy; and 15% from savings in infrastructure energy due to the
reduced server energy along with the relocation of servers into data centers with lower PUE
values.
4.5 Scenario Results
Server energy use is affected in all scenarios. Relative to the CT scenario, IM, BP, and HS
measures reduce server energy consumption by 4%, 18%, and 28% respectively. When HS
measures are applied to the IM and BP scenarios, server energy consumption is reduced an
additional 16% and 3% of the CT value, respectively. These results are shown in
Figure 32.
Storage energy use is only affected in the BP scenario where disks are assumed to be 25%
more efficient in 2020 than in the CT scenario. Therefore, storage energy is 25% less in the BP
and BP+HS scenarios relative to CT. Network energy is also only affected in the BP scenario
due to both port consolidation and port efficiency improvements. The result is a 38% reduction
in network energy consumption in BP and 40% in BP+HS scenarios relative to CT. This is
shown in Figure 33.
38
Figure 32. Server Electricity Use for All Scenarios
Figure 33. Network Electricity Use for Current Trends and Best Practices Scenarios.
Improved Operation, Hyperscale Shift, and IO+HS scenarios are identical to Current Trends. BP+HS
network electricity is slightly lower than BP alone due to the higher prevalence of network ports in
hyperscale data centers where the 25% wattage reduction is assumed to occur.
Infrastructure energy is affected in all scenarios, as shown in Figure 34. Infrastructure energy is
the product of the space type PUE and the space IT energy consumption, so reductions in both
of these areas compound for very large energy savings relative to the CT scenario. The HS
39
scenario has infrastructure energy savings of 46% relative to the current trends scenario. This is
due both to the reduction in server energy as well as the shift of this energy into hyperscale data
centers, which have a lower PUE than other space types. The IM scenario saves 22% of
infrastructure energy relative to CT due to both the reduction in IT energy and improvement in
PUE across smaller space types. IM and HS in conjunction save 54% of infrastructure energy.
Lastly, the BP and BP+HS scenarios have infrastructure energy savings of 64% and 75%
respectively due to the large decrease in server energy, additional decrease in storage and
network energy, and strong improvement in PUE across all space types.
Figure 34. Infrastructure Electricity Use for All Scenarios
Total data center energy consumption for all scenarios is shown in
Figure 35. Total electricity savings in 2020 for the IM, BP, and HS scenarios are 10%, 40%, and
25% respectively, relative to the CT scenario. When HS measures are applied to the IM and BP
scenarios an additional 20% and 6% of CT total energy is saved. Water savings are very similar
with scenario 2020 savings as follows: IM, 9%; BP: 39%; HS: 24%; IM+HS, 28% (18% more
than IM alone); and BP+HS, 45% (5% more than BP alone).
40
Figure 35. Total Electricity Consumption for All Scenarios
Figure 36. Water Consumption for All Scenarios
41
5 Indirect Energy Impacts
Data centers constitute a foundational component of the information and communication
technology (ICT) that provides the services—such as streaming media, email, internet content,
and electronic recordkeeping—now ubiquitous throughout much of the world.a Today, our
computers, mobile devices, sensors, and networks are more likely than ever to utilize
centralized information stored on a data center in the cloud. While the primary focus of this
report is to estimate the electricity directly consumed in data centers, it is important to take a
broader perspective and note that the services provided by data centers can profoundly affect
how—and how much—energyb is used elsewhere by society. For example, ICT can enable
existing products and services to become more efficient or create “e-substitutes” for physical
products. Other, higher-order effects occur when the introduction of ICT causes a change in
consumption or production elsewhere in the economy.
These implications for broader energy use are known as indirect energy impacts, and they
might either reduce or increase overall energy use (Figure 37). From an energy savings
standpoint, if the indirect energy impacts offset the direct consumption by data centers and
other ICT equipment, then the synergy between ICT and societal energy savings is positive; if,
instead, the introduction of ICT causes an amplifying effect in which overall energy consumption
increases, the synergy is negative. Characterizing this “net” impact of ICT deployment on
societal energy consumption has been of great interest, as evidenced by the emergence of an
ICT for Sustainability research community,62 two special issues in the Journal of Industrial
Ecology63 64 an OECD effort to link statistical indicators between the ICT and environment
research fields,65 work in “green computing” from the computer science field,66 and a variety of
other reports.67 68 This report chapter provides brief summary of the literature review and
interpretation found in Horner et al.69 on the indirect impacts of ICT.
There is, in fact, no consensus on the magnitude or even the sign of ICT’s indirect energy
impact. Generally in the positive synergy camp are Romm et al.70 and a series of reports
published by the American Council for an Energy Efficient Economy,71 72 who anticipate ICT-
enabled energy efficiency gains across broad sectors of the economy, and work commissioned
by the industry-sponsored Global e-Sustainability Initiative,73 which estimates a greenhouse gas
(GHG) abatement potential of 20% by 2030 due to ICT deployment. More cautionary is Rattle,74
who posits that higher-order effects are likely to swamp these sorts of energy savings
projections. Berkhout and Hertin75 argue for moving “beyond the dichotomy between pessimism
and optimism” to recognize that the relationship between ICT and energy impacts is “complex,
interdependent, deeply uncertain and scale-dependent.” Other literature reviews point to an
aThe term information technology (IT) has been used in this report to refer to the servers, network, and storage
equipment in the data center. In this chapter, ICT is used as a broader term encompassing data center IT
components, other network infrastructure, and a wide variety of end-use devices.
bResource usage and GHG emissions are, of course, concomitants of ICT energy consumption; while this report
focuses on energy, much of the literature includes these associated impacts.
42
ambiguous net impact or acknowledge that this complexity and uncertainty confound attempts
to verify a general belief that the net energy savings of ICT should be positive. 76 77 78 79
5.1 Energy Impact Taxonomy
Direct energy consumption, which in addition to the operational energy use estimated in this
report also includes the energy required to manufacture and dispose of ICT equipment, is likely
the simplest and ultimately the least important ICT energy effect80 although it is by no means
small. One particular issue is that embodied energy can dominate operational consumption for
some types of ICT equipment, such as mobile devices.78 However, the indirect energy effects
are likely to be of much greater magnitude,78 owing to the breadth of the various mechanisms by
which ICT services alter energy use. Table 8 breaks out individual effects, organizes them into a
taxonomy of increasing scope, and maps them to other terms used in the literature, while Figure
37 shows this taxonomy graphically.
Working from narrow to broad scope, ICT adoption first leads to efficiency in and substitution for
conventional products and services. Efficiency occurs when, for example, smart building
technology reduces air conditioning energy consumption by tailoring climate-control to the real-
time needs of building occupants. An example of substitution is the replacement of air travel
with teleconferencing. There is no guarantee, however, that the substituted ICT service will be
less energy intense than the conventional service it replaces, and even evaluation of simple
cases is not always straightforward.
Any energy reduction achieved through efficiency or substitution can be plagued by rebound
effects, in which expected gains are offset by induced additional consumption. Azevedo81,
Gillingham et al.82, and Borenstein83 provide comprehensive introductions to rebound effect
types. Rebound is typically broken into direct rebound, indirect rebound, and economy-wide
effects. Direct rebound effects are own-price-elasticity effects: as prices fall (due to
improvements in efficiency or productivity), substitution and income effects increase
consumption. For an ICT example, if an e-book is less costly than a conventional book, then
consumers might purchase more books. Alternatively, these savings could be spent on other
goods and services, which are indirect rebound effects. Indirect rebound effects result from
cross-price elasticity of demand for other products and services due to increased real consumer
income.
43
Table 8. Taxonomy of ICT Energy Effects from Horner et al.69
Scope of effect increases from top to bottom. The third column provides and example of each effect type
related to the deployment of Global Positioning System (GPS) Technology. Taxonomy expands on the
general taxonomy of rebound effects from Azevedo.81
Taxonomy summarized in this report Alternate taxonomies
Effect Scope GPS System Example Hilty84 Berkhout
& Hertin75 Williams91 Rattle74
Embodied
energy
Direct
Energy to produce a GPS
system
1st-
order
Direct
effects
ICT
infrastructure
and devices
Operational
energy
Energy to operate a GPS
system
Disposal
energy Energy to dispose of a GPS
system at end-of-life
Efficiency
Indirect:
Single-
service
More efficient traffic flow due
to GPS-enhanced routing 2nd-
order
Indirect
effects Applications Optimization
Substitution Replacement of paper maps Substitution
Direct rebound More travel due to lower
cost of traffic congestion
3rd-
order
Structural &
behavioral
effects
Effects on
economic
growth and
consumption
patterns
Induction
Indirect
rebound
Indirect:
Comple-
mentary
services
Energy consumed during
time saved by more efficient
travel
Supplement-
ation
Economy-wide
rebound
(Structural
change)
Indirect:
Economy-
wide
GPS enables autonomous
vehicles, causes growth of
intelligent transportation
system manufacturing
Creation
Systemic
Transformation
Indirect:
Society-
wide
Autonomous vehicles alter
patterns in where people
choose to live and work
Systemic
effects on
technology
convergence
& society
Economy-wide effects occur when the ICT introduction causes macroeconomic adjustments
across economic sectors. That is, the ICT industry can promote or inhibit growth in other sectors
of the economy, inducing structural changes that have energy use implications of their own. For
example, e-commerce is having broad effects on the logistics industry,85 including growth in
urban freight vehicle sales and changing patterns in distribution center floor space,86 increased
trucking and adoption of new pricing strategies by freight carriers87 and use of more specialized
packaging and a broader range of box sizes.88
Finally, transformational effects refer to the altering of human preferences and economic and
social institutions caused in part by the development of ICT.89 90 Historical examples include the
44
advent of the telephone and automobile, which heavily altered where and how people lived and
worked. We might conceive of a similar transformation (one of many possible ICT-enhanced
futures) in which the fundamental constraints on where people live and work continue to loosen:
e-commerce and home delivery make proximity to traditional retail outlets less important,
seamless telework results in less commuting, and driverless vehicles allow for more productive
use of the commuting time.
Figure 37. Taxonomy of Energy Effects from Adoption of ICT, from Horner et al.69
Red effects increase energy use, blue effects decrease energy use, and shading intensity decreases as
effect scope increases.
As noted by Börjesson Rivera et al.,76 the existing literature uses several different sets of terms
for this hierarchy of effects, collected in the right half of Table 8. ICT energy effects are
frequently grouped into first-order impacts due to direct consumption, second-order effects
resulting from process changes, such as efficiency, and third-order effects due behavioral and
economic changes.75,84 Williams91 adds a fourth level, essentially breaking third-order effects
into rebound effects and broader systemic change.
Indirect
Rebound
Direct
Rebound
Subs tu on
ICT
Equipment
Direct
Consump on
Effic ienc y
Structural
Economic
Changes
Systemic
Transforma on
Net Energy Use
+
Scope of Impact
Single
service
Complementary
services
Economy
and
societywide
Subs tu on
Disposal
Energy
Opera onal
Energy
Embodied
Energy
Direct
45
5.2 Energy Impact Estimation
The literature on characterizing these indirect impacts can generally be divided into two broad
types. One body of literature typically assesses the impact of a single service, such as e-
commerce, telecommuting, or smart buildings. Such analyses often use lifecycle assessment
(LCA),92 93 modeling & simulation,94 or case studies95 96 and generally address only direct
consumption, efficiency, substitution, and occasionally direct rebound. A separate body of
literature focuses on the higher-order indirect energy impacts brought about by ICT-induced
changes to complementary services, the broader economy, and societal systems. These
studies typically rely on econometric analysis of macroeconomic indicators,97 scenario
analysis,98 99 or anecdotal evidence.100
Importantly, neither type of study uniformly finds a positive or negative effect: results from both
are highly dependent on modeling choices and assumptions used in scoping the study. For
example, the energy savings reported in LCA studies of e-commerce for book retail range from
negative 500% (i.e., a 5x increase in energy use in the e-commerce case) to nearly 50% (a
reduction in energy use by half).69 This variation results from differences in the system boundary
and from sensitivity to assumptions, including population density, freight mode, product return
rate, proportion of multipurpose trips, and packaging type. Case studies often show energy
savings in specific deployment scenarios and provide valuable lessons on how to deploy ICT in
such a way that energy savings are attained; however, scalability and higher-order effects
remain uncertain in such work. Macroeconomic studies also show highly variable results.67
Thus, while both conceptual discussion and analytical modeling of ICT energy and
environmental impacts have been occurring for at least two decades, the jury is still out on the
net effects of ICT adoption for several reasons. First, the complexity and variability of ICT
deployment schemes makes it difficult to isolate a standard implementation to analyze and to
compare study results. Second, the lack of empirical data on how human users interact with ICT
systems hinders the ability to assess actual, instead of potential, energy effects. Third, the
difficulties in disentangling the causes of interconnected effects lead to a tendency to fall back
on theory—and on modeling exercises that conform to these theories. Finally, as the impact
scope increases up the effect taxonomy (Table 8), the potential effect’s magnitude and
uncertainty increase dramatically.
The current state of understanding can be summarized with three related statements: the
technical potential of ICT net energy savings is likely positive; the sign and magnitude of
realized net energy savings are highly sensitive to the parameters that characterize the ICT
deployment and are not guaranteed; and, finally, the actual net energy effect is unclear and
difficult to assess, especially when higher-order impacts are considered.
5.3 Pathway Forward
Just as implementation of best practices in data center design and operation has the potential to
drastically reduce direct energy consumption (Figure 35), optimizing the manner in which we
integrate ICT into our lives can have a large impact on overall energy consumption. During the
first decade of the century, data center energy consumption grew rapidly (Figure 23). IT
46
managers focused on service provisioning, with the power bill being a much lesser concern.
However, the industry is now converging on a paradigm of virtualization and centralization that
has both business and energy co-benefits, and such reductions in direct energy use should
continue to be pursued.
The broader evolution of ICT is perhaps on a similar path: new systems and services are being
developed rapidly without much consideration of energy impacts, and as a result it seems likely
that ICT services are often deployed in a way that does not achieve their full potential to achieve
energy savings. However, it also seems likely that more optimal deployment plans—those that
create energy savings while maintaining the value of the service—exist, and more focus on
characterizing these “system optima” is warranted. 101
The danger in waiting to identify these deployment plans is that society-wide systems,
structures, and habits that become entrenched can be much more difficult to alter. A server has
a typical lifetime less than five years; the economic and social infrastructures built out through
new ICT services can last much longer. Thus, the important role of analysis in this area is to
identify the important drivers of ICT indirect energy impacts and gather data on actual, rather
than potential, energy savings, so that these results can inform both public policy and private
decision making on the implementation and use of ICT. For this reason, the field would benefit
from more focus on empirical case studies and on understanding the behavioral aspects of how
various stakeholders use ICT services in practice. Additional work on characterizing uncertainty
in energy effect estimates would also benefit discussions in this area.
6 Future Work
Estimating U.S. data center energy use requires developing inputs and assumptions for an
industry with rapidly evolving technologies and limited publically available energy use data,
which ultimately limits the potential scope of analysis. Through the challenges of developing
data center energy use estimates for this report, additional areas of research were identified that
could improve future growth estimates in data center energy consumption and the potential
impacts for specific efficiency efforts. Below are key areas identified that warrant future
research.
6.1 Server Utilization and Power Proportionality
Due to the limited data on utilization rates for servers in U.S. data centers, this study
generalizes server utilization using a single average (per space type), with no information about
the actual distribution of utilization over the average year. This generalization prevents
distinguishing between a server that is run at 40% utilization constantly and one that is run at
80% utilization half of the time and idled the other half of the time. Additionally, while this study
assumes a linear relationship between utilization and power consumption (i.e., the “scaling
curve”) of servers, the actual relationship is generally nonlinear18 and therefore there is a loss of
accuracy in modeling server energy use at the average utilization level. Without the ability to
model nonlinear scaling curves, it then becomes difficult to understand the impacts of certain
47
efficiency measures, such as the targeted lowering of idle state power consumption (e.g.
Emerson Network Power’s proposed 10 Minus standard102) or powering down servers during
idle times. As consolidation efforts like virtualization are increasing server utilization levels
across the industry, it is even more important to understand how future energy savings
opportunities associated with increasing dynamic ranges and low-power idle states are going to
be affected.
6.2 Workload Variation
In addition to understanding server utilization over time, further efforts should be made to
understand the distribution of various types of server workloads and their associated hardware
requirements. This will help to quantify opportunities for energy savings associated with
optimizing hardware for specific workloads as opposed to using a “one-size-fits-all” approach
prevalent across large data centers today. This could include using single-socket designs and
workload-optimized processors (RISC, FPGA, GPUs, etc.). Understanding workload variations
among different servers within data centers would also assist in identifying strategies to further
increase overall data center utilization loads and increase cooling efficiency by creating the
opportunity to provide server-specific cooling demand.
6.3 Barriers to Hyperscale Shift
While there has been significant growth in hyperscale data centers, this report shows that a
significant portion of servers are still expected to reside in small room or closet data centers.
Better understanding the different barriers, including technical, legal, and security barriers, that
are preventing movement to colocation or to the cloud can help drive solutions that increase the
shift to large data centers and tailor energy efficiency strategies for the small data centers that
remain.
6.4 Beyond PUE
There is a need in the data center industry for performance metrics that better capture the
efficiency of a given data center. The limitations of PUE, the most commonly discussed metric
of efficiency, are generally understood,103 but a key issue it that PUE only measures the
efficiency of the building infrastructure supporting a given data center and indicates nothing
about the efficiency of the IT equipment itself. Metrics that capture the functionality of the data
center (e.g. amount of computations it performs) and relate that to energy use can help industry
better understand where progress has been made and where there are opportunities to reduce
energy use. Initial efforts to accomplish this include The Green Grid’s Data Center Productivity
(DCP) and Data Center energy Productivity (DCeP),104 the Uptime Institute and McKinsey’s
Corporate Average Data center Efficiency (CADE),105 and JouleX’s Performance per Watt
(PPW),106 but PUE is still the dominant metric broadly observed in the data center industry.
6.5 Beyond 2020
The significant energy efficiency improvements in the design and operation of data centers over
the past decade have allowed U.S. data center energy use to remain nearly constant while
48
simultaneously meeting a drastic increase in demand for data center services. However, the
data available at the time of this study limited the scope of future projection to 2020. The key
efficiency strategies identified in this report, improved PUE, increased server utilization rates,
and better power proportionality all have theoretical and practical limits and the current rate of
improvement indicates that these limits may be reached in the not too distant future. The
potential for data center services, especially from a global prospective, are still in a fairly
nascent stage and future demand could continue to increase after our current strategies to
improve energy efficiency have been maximized. Understanding if, and when, this transition
may occur, and the ways in which data centers can minimize their costs and environmental
impacts under such a scenario, is an important direction for future research. This report
highlights the success of the data center industry to stabilize electricity demand, but further
investigation and technological breakthroughs in energy efficiency across the ICT equipment
spectrum will be needed to insure that success is not simply a plateau before an increase in
electricity demand resumes at a rate proportional to future growth of data center services.
49
References
1 Brown, R., Masanet, E., Nordman, B., Tschudi, W., Shehabi, A., Stanley, J., Koomey, J.,
Sartor, D., Chan, P., Loper, J., Capana, S., Hedman, B., Duff, R., Haines, E., Sass, D., and A.
Fanara. (2007). Report to Congress on Server and Data Center Energy Efficiency: Public Law
109-431. Lawrence Berkeley National Laboratory, Berkeley, California. LBNL-363E.
2 Masanet, E, Brown, R E, Shehabi, A, Koomey, J G, and B Nordman (2011). “Estimating the
Energy Use and Efficiency Potential of U.S. Data Centers. Proceedings of the IEEE, Volume 99,
Number 8.
3 Koomey, J.G. (2011). Growth in Data Center Electricity Use 2005 to 2010. Analytics Press,
Oakland, California. http://www.analyticspress.com/datacenters.html
4 Upton, F (2015). North American Energy Security and Infrastructure Act of 2015. H.R. 8, 114th
Congress. https://www.congress.gov/bill/114th-congress/house-bill/8
5 Koomey, Jonathan. 2008. Worldwide electricity used in data centers. Environmental Research
Letters. vol. 3, no. 034008. September 23. [http://stacks.iop.org/1748-9326/3/034008]
6 Koomey, Jonathan G. 2007. Estimating Total Power Consumption by Servers in the U.S. and
the World. February 15.
http://www.mediafire.com/file/exywo1hf6ionskw/AMDserverpowerusecomplete-final.pdf.
7 IDC Worldwide Quarterly Server Tracker:
http://www.idc.com/tracker/showproductinfo.jsp?prod_id=7
8 International Data Corporation (IDC). 2015. IDC’s Worldwide Quarterly Server Shipment
Tracker, 2010-2018, Framingham, MA, March.
9 International Data Corporation (IDC). 2015. IDC’s Worldwide Quarterly Disk Storage Systems
Tracker, 2010-2019, Framingham, MA, March.
10 International Data Corporation (IDC). 2015. IDC’s Worldwide Quarterly Data Center
Networks, 2008-2019, Framingham, MA, March.
11 International Data Corporation (IDC). 2014. IDC's Worldwide Quarterly Server Tracker –
Installed Base, 2006-2018. Framingham, MA: IDC. December.
12 Dietrich, J. 2014 ITIC Analysis of SERT Worklet Results. Information Technology Industry
Council. Available at:
https://www.energystar.gov/sites/default/files/specs/ITI%20Analysis%20of%20SERT%20Data.p
df
13 International Data Corporation (IDC). 2015. Personal communication with Lidice Fernandez,
Program Vice President, WorldWide Tracker Research, September 9.
50
14 The Green Grid, Data Centre Life Cycle Assessment Guidelines. White Paper #45, v2, 2012.
Available at:
http://www.thegreengrid.org/~/media/WhitePapers/WP45v2DataCentreLifeCycleAssessmentGui
delines.pdf
15 SPEC (2015). SPECpower_ssj2008 Results. Downloaded September 10, 2015 from
https://www.spec.org/power_ssj2008/results/
16 Van Heddeghem, W., Lambert, S., Lannoo, B., Colle, D., Pickavet, M., & Demeester, P.
(2014). Trends in worldwide ICT electricity consumption from 2007 to 2012. Computer
Communications, 50, 64-76.
17 NRDC and WSP (2012). The Carbon Emissions of Server Computing for Small- to Medium-
Sized Organizations: A Performance Study of On-Premise vs.The Cloud. WSP Environment &
Energy, LLC and Natural Resources Defense Council. October 2012.
18 Barroso, L. A., Clidaras, J., & Hölzle, U. (2013). The datacenter as a computer: An
introduction to the design of warehouse-scale machines. Synthesis lectures on computer
architecture, Morgan Claypool Publishers
19 Hylick, A., Ripduman, S., Rice, A., Jones, B. 2008. An Analysis of Hard Drive Energy
Consumption. University of Cambridge, Computer Laboratory, October. Available at:
https://www.cl.cam.ac.uk/~acr31/pubs/hylick-harddrive2.pdf
20 ASHRAE (2015). Data Center Storage Equipment – Thermal Guidelines, Issues, and Best
Practices. Technical Committee 9.9.
21 Cadmus (2015). New York State Data Center Market Characterization. New York State
Energy Research and Development Authority (NYSERDA) Report Number 15-06. October
2015.
22 Reinsel, David (2010). A Plateau in Sight for the Rising Costs to Power and Cool the World’s
External Storage? IDC Opinion, IDC#225016. September 2010.
23 Lanzisera, S., Nordman, B., & Brown, R. E. (2012). Data network equipment energy use and
savings potential in buildings. Energy Efficiency, 5(2), 149-162.
24 Reviriego, P., Maestro, J. A., & Larrabeiti, D. (2010). Burst transmission for energy-efficient
ethernet. Internet Computing, IEEE, 14(4), 50-57.
25 Dudkowski, D., Hasselmeyer, P. Energy-Efficient Networking in Modern Data Centers. In
Konstantinos Samdanis, Peter Rost, Andreas Maeder, Michaela Meo, Christos Verikoukis:
Green Communications: Principles, Concepts and Practice, John Wiley & Sons, 2015. ISBN
978-1-118-75926-4
51
26 Bailey, M., M. Eastwood, T Grieser, L. Borovick, V. Turner, and R.C. Gray. 2007. Special
Study: Data Center of the Future. New York, NY: IDC. IDC #06C4799. April.
27 Villars, Richard L (2014). “U.S. Datacenter Census and Construction 2014-2018 Forecast:
Realigning Workloads, Managing Obsolescence, and Leveraging Hyperscale”. IDC #252712.
28 Venture Outsource, 2015. ODM Quanta focus on white boxes bypasses Dell, HP, Cisco
traditional model for OEMs. VentureOutsource.com:
https://www.ventureoutsource.com/contract-manufacturing/focus-odm-quanta-it-shift-cloud-
infrastructure-leaving-dell-hp-traditional
29 IDC (2014). The Digital Universe of Opportunities: Rich Data and the Increasing Value of the
Internet of Things. April 2014, sponsored by EMC. http://idcdocserv.com/1678.
30 International Data Corporation (IDC). 2015. Personal communication with Rick Villars,
Program Vice President, Data Center and Cloud Research, November 12.
31 The Green Grid datacenter power efficiency metrics: PUE and DCiE. Available at
http://www.thegreengrid.org/sitecore/content/Global/Content/white-papers/The-Green-Grid-
Data-Center-Power-Efficiency-Metrics-PUE-and-DCiE.aspx.
32 Tschudi, William, Tengfang Xu, Dale Sartor, and Jay Stein. 2003. High Performance Data
Centers: A Research Roadmap. Berkeley, CA: Lawrence Berkeley National Laboratory. LBNL-
53483. http://hightech.lbl.gov/documents/DataCenters_Roadmap_Final.pdf.
33 Greenberg, Steve, Evan Mills, Bill Tschudi, Peter Rumsey, and Bruce Myatt. 2006. Best
Practices for Data Centers: Lessons Learned from Benchmarking 22 Data Centers.
Proceedings of the ACEEE Summer Study on Energy Efficiency in Buildings in Asilomar, CA.
ACEEE, August. Vol 3, pp 76-87. http://eetd.lbl.gov/emills/PUBS/PDF/ACEEE-datacenters.pdf.
34 C. Malone and C. Belady. Metrics to characterize datacenter & IT equipment energy use. In
Proceedings of the Digital Power Forum, Richardson, TX, September 2006.
35 Sullivan, A. 2010. Energy Star for Data Centers. Green Grid Forum. February 4, 2010
36 Uptime Institute, Important to recognize the dramatic improvement in data center
efficiency.http://blog.uptimeinstitute.com/2012/09/important-to-recognize-the-dramatic-
improvement-in-data-center-efficiency/
37 Cheung, I. H., Greenberg, S., Mahdavi, R., Brown, R., & Tschudi, W. (2014, August). Energy
Efficiency in Small Server Rooms: Field Surveys and Findings. Proceedings the 2014 ACEEE
Summer Study on Energy Efficiency in Buildings. LBNL- 6952E
38 Google 2015. “Our energy-saving data centers.” Accessed December 2, 2015.
http://www.google.com/about/datacenters/efficiency/internal/index.html#measuring-efficiency
39 Gelber, R. 2012. “Facebook showcases green datacenter.” HPCwire. Accessed December 2,
2015. http://www.hpcwire.com/hpcwire/2012-04-
26/facebook_showcases_green_datacenter.html
52
40 Masanet, E., Shehabi, A., Ramakrishnan, L., Liang, J., Ma, X., Walker, B., & Mantha, P.
(2013). The Energy Efficiency Potential of Cloud-Based Software: A US Case Study. Lawrence
Berkeley National Laboratory, Berkeley, California.
41 Shehabi, A., Masanet, E., Price, H., Traber, K., Horvath, A., and W.W. Nazaroff. 2011. “Data
Center Design and Location: Consequences for Electricity Use and Greenhouse-Gas
Emissions.Building and Environment, Volume 46, Issue 5.
42 Koomey, J.G., Berard, S., Sanchez, M. and Wong, H., 2011. Implications of historical trends
in the electrical efficiency of computing. Annals of the History of Computing, IEEE, 33(3), pp.46-
54.
43 Koomey, Jonathan, and Samuel Naffziger. 2015. "Efficiency's brief reprieve: Moore's Law slowdown
hits performance more than energy efficiency." In IEEE Spectrum. April. pp.
[http://spectrum.ieee.org/computing/hardware/moores-law-might-be-slowing-down-but-not-energy-
efficiency]
44 The Green Grid. 2011. Water Use Effectiveness: A Green Grid Data Center Sustainability
Metric. White Paper #35.
45 P. Torcellini, N. Long, and R. Judkoff. Consumptive Water Use for U.S. Power Production.
2003, NREL/TP-550- 33905, http://www.nrel.gov/docs/fy04osti/33905.pdf
46 Miller, R. 2009. Data Centers Move to Cut Water Waste. Data Center Knowledge. Available
at: http://www.datacenterknowledge.com/archives/2009/04/09/data-centers-move-to-cut-water-
waste
47 Fitzgerald, D. 2015. Data Centers and Hidden Water Use. The Wall Street Journal. Available
at: http://www.wsj.com/articles/SB10007111583511843695404581067903126039290
48 Federal Data Center Consolidation Initiative (FDCCI), 2014. GSA FDCCI Inventory Data
Fields. Available at: https://cio.gov/wp-content/uploads/downloads/2012/10/FDCCI-Inventory-
Data-Fields-Release-2.pdf. Accessed: March 25, 2014.
49 Boyd, A. 2015. Experts: Data center consolidation goals not aggressive enough. Federal
Times.com. Available at: http://www.federaltimes.com/story/government/it/data-
center/2015/02/17/data-center-consolidation-goals-not-aggressive/23556995/
50 Top500. 2016. Top500 List – November 2015. http://www.top500.org/lists/2015/11/
51 Kaplan, JM., Forrest, W., Kindler, N. 2008. Revolutionizing Data Center Efficiency McKinsey
and Company.
http://www.mckinsey.com/clientservice/bto/pointofview/pdf/revolutionizing_data_center_efficienc
y.pdf
52 The Uptime Institute estimate; https://uptimeinstitute.com/research-
publications/asset/comatose-server-savings-calculator
53
53 Koomey, J., Taylor J. 2015. New data supports finding that 30 percent of servers are
‘Comatose’ , indicating that nearly a third of capital in enterprise data centers is wasted.
http://anthesisgroup.com/wp-content/uploads/2015/06/Case-
Study_DataSupports30PercentComatoseEstimate-FINAL_06032015.pdf
54 McMillian, R. 2015. Zombie Servers: They’re Here and Doing Nothing but Burning Energy.
The Wall Street Journal. September 13, 2015. http://www.wsj.com/articles/zombie-servers-
theyre-here-and-doing-nothing-but-burning-energy-1442197727
55 Delforge, P., & Whitney, J. 2014. Issue Paper: Data Center Efficiency Assessment Scaling up
Energy Efficiency Across the Data Center Industry: Evaluating Key Drivers and Barriers. Natural
Resources Defense Council (NRDC).
56 Ourghanlian, B. 2010. Improving Energy Efficiency: And End User Perspective. The Green
Grid. The Green Grid Technical Form 2010. Available at:
http://www.thegreengrid.org/~/media/EMEATechForums2010/Improving%20Energy%20Efficien
cy%20-%20An%20End%20User%20Perspective_Paris.pdf?lang=en
57 United EIF, Inc. 2015. Advanced Configuration and Power Interface Specification. Available
at: http://www.uefi.org/sites/default/files/resources/ACPI_6.0.pdf
58 Lo, D., Cheng, L., Govindaraju, R., Ranganathan, P., & Kozyrakis, C. (2015, June). Heracles:
improving resource efficiency at scale. In Proceedings of the 42nd Annual International
Symposium on Computer Architecture (pp. 450-462). ACM.
59 Masanet, E., Shehabi, A., Ramakrishnan, L., Liang, J., Ma, X., Walker, B., Hendrix, V., and P.
Mantha (2013). The Energy Efficiency Potential of Cloud-Based Software: A U.S. Case Study.
Lawrence Berkeley National Laboratory, Berkeley, California.
60 Heller, B., Seetharaman, S., Mahadevan, P., Yiakoumis, Y., Sharma, P., Banerjee, S., &
McKeown, N. (2010, April). ElasticTree: Saving Energy in Data Center Networks. In NSDI (Vol.
10, pp. 249-264).
61 Masanet, E., A. Shehabi, L. Ramakrishnan, J. Liang, X. Ma, B. Walker, V. Hendrix, and P.
Mantha (2013). The Energy Efficiency Potential of Cloud-Based Software: A U.S. Case Study.
Lawrence Berkeley National Laboratory, Berkeley, California.
62 Hilty, L.M., Aebischer, B. (Eds.), 2015. ICT Innovations for Sustainability, Advances in
Intelligent Systems and Computing. Springer International Publishing, Cham.
63 Masanet, E.R., Matthews, H.S. (Eds.), 2010. Environmental Applications of Information and
Communication Technology [special issue]. J. Ind. Ecol. 14, 685–862.
64 Rejeski, D. (Ed.), 2002. E-Commerce, the Internet, and the Environment [special issue]. J.
Ind. Ecol. 6, 1–161.
65 Roberts, S., 2009. Measuring the Relationship between ICT and the Environment.
Organization for Economic Co-operation and Development.
http://www.oecd.org/sti/43539507.pdf
54
66 Auweter, D. Kranzlmüller, A. Tahamtan, A. M. Tjoa, D. Hutchison, T. Kanade, J. Kittler, J. M.
Kleinberg, F. Mattern, J. C. Mitchell, M. Naor, O. Nierstrasz, C. Pandu Rangan, B. Steffen, M.
Sudan, D. Terzopoulos, D. Tygar, M. Y. Vardi, and G. Weikum, Eds., ICT as Key Technology
against Global Warming, ser. Lecture Notes in Computer Science. Berlin, Heidelberg: Springer
Berlin Heidelberg, 2012, vol. 7453, ISBN: 978-3- 642-32605-9. [Online]. Available:
http://link.springer.com/10.1007/978-3-642-32606-6
67 Erdmann, L., Hilty, L., Goodman, J., Arnfalk, P., 2004. The Future Impact of ICTs on
Environmental Sustainability (Technical Report No. EUR 21384 EN). Institute for Prospective
Technological Studies, Seville, Spain.
68 Jørgensen, M.S., Andersen, M.M., Hansen, A., Wenzel, H., Thoning, T., Pedersen, U.J.,
Falch, M., Rasmussen, B., Olsen, S.I., Willum, O., 2006. Green Technology Foresight about
environmentally friendly products and materials. Environ. Prot. Agency Den.
http://www2.mst.dk/udgiv/publications/2006/87-7052-216-2/pdf/87-7052-217-0.pdf.
69 Horner, N., Shehabi, A., Azevedo, I. Known unknowns: Indirect energy effects of information
and communication technology. (In Review)
70 Romm, J., Rosenfeld, A., Herrmann, S., 1999. The internet economy and global warming: A
scenario of the impact of e-commerce on energy and the environment. Cent. Energy Clim.
Solut. Dec. Httpwww Cool-Co. Orgenergycecs Cfm.
71 Laitner, J.A., Ehrhardt-Martinez, K., 2008. Information and Communication Technologies: The
Power of Productivity (E081). American Council for an Energy-Efficient Economy, Washington,
D.C.
72 Elliott N, Molina, M., and Trombley, D. 2012, A defining framework for intelligent efficiency
(Washington, DC: American Council for an Energy-Efficient Economy) Online:
http://aceee.org/sites/default/files/publications/researchreports/e125.pdf
73 Accenture. 2015. SMARTer2030: ICT Solutions for 21st Century Challenges Online:
http://smarter2030.gesi.org/downloads/Full_report2.pdf
74 Rattle, R., 2010. Computing Our Way to Paradise?: The Role of Internet and Communication
Technologies in Sustainable Consumption and Globalization. Rowman & Littlefield.
75 Berkhout, F., Hertin, J., 2004. De-materialising and re-materialising: digital technologies and
the environment. Futures 36, 903–920. doi:10.1016/j.futures.2004.01.003
76 Börjesson Rivera, M., Håkansson, C., Svenfelt, Å., Finnveden, G., 2014. Including second
order effects in environmental assessments of ICT. Environ. Model. Softw. 56, 105–115.
doi:10.1016/j.envsoft.2014.02.005
77 Erdmann, L., Hilty, L.M., 2010. Scenario Analysis: Exploring the Macroeconomic Impacts of
Information and Communication Technologies on Greenhouse Gas Emissions. J. Ind. Ecol. 14,
826–843. doi:10.1111/j.1530-9290.2010.00277.x
55
78 Koomey, J.G., Matthews, H.S., Williams, E., 2013. Smart Everything: Will Intelligent Systems
Reduce Resource Use? Annu. Rev. Environ. Resour. 38, 311–343. doi:10.1146/annurev-
environ-021512-110549
79 Yi, L., Thomas, H.R., 2007. A review of research on the environmental impact of e-business
and ICT. Environ. Int. 33, 841–849. doi:10.1016/j.envint.2007.03.015
80 Allenby, B., Unger, D., 2001. Information technology impacts on the US energy demand
profile. RAND Corp.
81 Azevedo, I.M.L., 2014. Consumer End-Use Energy Efficiency and Rebound Effects. Annu.
Rev. Environ. Resour. 39, 393–418. doi:10.1146/annurev-environ-021913-153558
82 Gillingham, K., Rapson, D., Wagner, G., 2015. The rebound effect and energy efficiency
policy.
83 Borenstein, S., 2013. A microeconomic framework for evaluating energy efficiency rebound
and some implications. National Bureau of Economic Research.
84 Hilty, L.M., Arnfalk, P., Erdmann, L., Goodman, J., Lehmann, M., Wäger, P.A., 2006. The
relevance of information and communication technologies for environmental sustainability – A
prospective simulation study. Environ. Model. Softw. 21, 1618–1629.
doi:10.1016/j.envsoft.2006.05.007
85 Hesse, M., 2002. Shipping news: the implications of electronic commerce for logistics and
freight transport. Resour. Conserv. Recycl. 36, 211–240.
86 Harrington, D., 2015. From first mile to last mile: Global industrial & logistics trends. Colliers
International.
87 Shorr Packaging Corp, 2015. The Amazon Effect: Impacts on Shipping and Retail Available
at: http://www.shorr.com/packaging-news/2015-06/amazon-effect-impacts-shipping-and-retail
(accessed 11.30.15).
88 Mohan, A.M., 2014. E-commerce packaging pitfalls & opportunities. Packag. World.
89 Greening, L.A., Greene, D.L., Difiglio, C., 2000. Energy efficiency and consumption -- the
rebound effect -- a survey. Energy Policy 28, 389–401.
90 Plepys, A., 2002. The grey side of ICT. Environ. Impact Assess. Rev. 22, 509–523.
91 Williams, E., 2011. Environmental effects of information and communications technologies.
Nature 479, 354–358. doi:10.1038/nature10682
92 Weber, C. et al. Life cycle comparison of traditional retail and e-commerce logistics for
electronic products: A case study of buy.com. Green Des. Inst. Carnegie Mellon Univ. (2008). at
http://www.ce.cmu.edu/~greendesign/research/Buy_com_report_final_030209.pdf
56
93 Shehabi, A., Walker, B. & Masanet, E. The energy and greenhouse-gas implications of
internet video streaming in the United States. Environ. Res. Lett. 9, 054007 (2014).
94 Kitou, E. & Horvath, A. External air pollution costs of telework. Int. J. Life Cycle Assess. 13,
155–165 (2008).
95 Henderson, P. & Waitner, M. Real-time energy management: A case study of three large
commercial buildings in Washington, D.C. (Natural Resources Defense Council, 2013). at
http://www.nrdc.org/business/casestudies/files/tower-companies-case-study.pdf
96 Seidel, S. and Ye, J. 2012. Leading by example: using information and communication
technologies to achieve Federal sustainability goals (Center for Climate and Energy Solutions)
Online: http://www.c2es.org/docUploads/federal-sustainability-ict.pdf
97 Laitner, J. A. & Ehrhardt-Martinez, K. Information and Communication Technologies: The
Power of Productivity. (American Council for an Energy-Efficient Economy, 2008). at
http://aceee.org/sites/default/files/publications/researchreports/E081.pdf
98 Baer, W. S., Hassell, S. & Vollaard, B. A. Electricity requirements for a digital society. (RAND
Corporation, 2002).
99 Hilty, L. M. et al. The relevance of information and communication technologies for
environmental sustainability – A prospective simulation study. Environ. Model. Softw. 21, 1618–
1629 (2006).
100 Romm, J., Rosenfeld, A. & Herrmann, S. The internet economy and global warming: A
scenario of the impact of e-commerce on energy and the environment. Cent. Energy Clim.
Solut. Dec. Httpwww Cool-Co. Orgenergycecs Cfm (1999). at
http://www.fraw.org.uk/files/tech/romm_1999.pdf
101 Laitner, J.A., McDonnell, M.T., and Keller, R.M. 2015. ICT-Enabled Intelligent Efficiency:
Shifting from device-specific approaches to system optima. Online: http://cda.iea-
4e.org/document/11/ict-enabled-intelligent-efficiency-shifting-from-device-specific-approaches-
to-system-optima
102 Pouchet, J. 2016. Time for a Server Idle Performance Standart – Introducing 10 Minus.
Emersion Network Power Blog. Available at:
http://blog.emersonnetworkpower.com/efficiency/time-for-a-server-idle-performance-standard/
103 Horner, N., Azevedo, I. 2016. Power usage effectiveness in data centers: overloaded and
underachieving. The Electricity Journal, Volume 29, Issue 4, May 2016, Pages 61-69
104 Azevedo, D., Rawson, A. 2008. Measuring Data Center Productivity. Metrics and
Measurements Work Group. The Green Grid. Available at:
http://www.thegreengrid.org/~/media/TechForumPresentations2008/Measuring_Data_Center_P
roductivity.pdf?lang=en
57
105 Kaplan, J., Forrest, W., Kindler, N. 2008. Revolutionizing Data Center Energy Efficiency.
McKinsey & Company. Available at: http://www.sallan.org/pdf-
docs/McKinsey_Data_Center_Efficiency.pdf
106 Cappiello, C., Chen, D., Ferreria, A.M., Henis, R., Jiang, T., Kat, I., Kipp, A., Liu, J., Sotnikov,
D., Vitali, M. 2011. Usage Centric Green Performance Indicators. Available at:
http://www.sigmetrics.org/sigmetrics2011/greenmetrics/green11_Kat_email.pdf
... type of services. However, data centers generally can be categorized into "internal" data center (representing traditional facilities servicing businesses and institutions) and "service provider" data centers (addressing facilities provide special services like social media and communications) [22]. The International Data Corporation (IDC) sorted data centers based on a combination of former classifications, tiers, and size of the facility. ...
... In terms of distinct infrastructure, a data center can be essentially disaggregated into: 1) its power infrastructurewhich includes power subsystems and Uninterrupted Power Supply (UPS); 2) its IT infrastructureembracing servers, external storage, and networking/ connection where there are two types of servers ("1 S" with single processor socket and "2 S+" with two or more processor sockets) and two types of storage, hard disk drives and solid state drives [22]; 3) its cooling infrastructurecontaining air-cooled systems and/or liquid-cooled systems; and 4) its peripheral systemssuch as backup generator and fire suppression. Fig. 2a shows that each of these subcategories are associated with some amount of energy consumption depending on the type of the utilized systems. ...
... Fig. 2a shows that each of these subcategories are associated with some amount of energy consumption depending on the type of the utilized systems. For instance, the servers, external storage, and networking are the primary energy consumers in the IT infrastructure [22]. Water consumption has recently been gaining some attention beyond the electricity demand of the data center. ...
Article
Over the last decade, the demand for data center and network services has increased dramatically. To meet this demand, global average rack power density has risen from 2.4 kW/rack in 2011 to 8.4 kW/rack in 2020 with the aid of technological advancements. About 36% of global data centers have racks above 30 kW/rack. Average Power Usage Effectiveness (PUE) dropped from 2.5 in 2007 to 1.65 in 2013, but has been almost flat since then, at 1.59 in 2020. The depletion of finite fossil fuel reservoirs, adverse impact of greenhouse gas emissions on global warming, recyclability of electronic waste, and the rising cost of energy are pushing the datacom industry toward more energy-efficient and sustainable technologies including renewable energy, highly efficient elec- tronics cooling, waste heat recovery, passive cooling systems, and energy storage, which are broadly reviewed in this paper. Use of renewable energy resources through power purchase agreements rose by 346% from 2.4 GW in 2017 to 10.7 GW in 2020. Air-side economizers can reduce PUE by 30–50% and water-side economizers by 10–30%. The server/rack outlet is an ideal spot for waste heat recovery and can be coupled with a multi-stage heat pump to save significant energy if coupled to a district heating system. Active and passive deployment of thermal energy storage can save more than 20% of cooling energy and reduce temperature fluctuations by more than 60%. Edge computing, decentralization, and virtualization can lead to lower latency, higher scalability, and reliability during failure
... According to [6] data centers are using 2% of the total US electricity demands, and by using energy efficient techniques 80% of savings can be made. A report published by the U.S. Department of Energy estimated that data centers operating in U.S. consumed approximately 70 billion kilowatt-hours in 2014 [7,8]. The energy consumption of U.S. data centers reached 200 billion kilowatt-hours by the end of 2020 [8]. ...
... A report published by the U.S. Department of Energy estimated that data centers operating in U.S. consumed approximately 70 billion kilowatt-hours in 2014 [7,8]. The energy consumption of U.S. data centers reached 200 billion kilowatt-hours by the end of 2020 [8]. ...
... It should be noted that doing these extensive analyses is not a purely positive thing. The computing power required results in large servers that run for hours, resulting in high energy consumption and the associated environmental impact [254]. For a big data system in the ID sector, it will be necessary to look at how these effects can be prevented or compensated. ...
Article
Full-text available
It was last estimated that in 2020, data centers comprised approximately 2% of total US electricity consumption, with an estimated annual growth rate of 4%. As our country increasingly relies on information technology (IT), our data centers (DCs) will need to increase their energy efficiency (EE) to stabilize their energy consumption. The task of studying EE in DCs is complicated by the interconnected nature of humans and mission-critical technical systems. Moreover, the literature tends to focus on technology solutions such as improvements to IT equipment, cooling infrastructure, and software, without addressing organizational and psychological drivers. Our research demystifies the complex interactions between humans and DCs, by asking What non-technical barriers impede EE investment decision-making and/or implementing energy management strategies? To begin to answer this question, we perform a literature review of 86 resources, ranging from peer-reviewed journal publications to handbooks. We also consider related fields such as organizational behavioral management and energy intensive buildings. We develop a public Zotero library, perform content coding, and complete a rudimentary network analysis. Our findings from the literature review suggest that (1) technological solutions are abundant in the literature but fall short of providing practical guidance on the pitfalls of implementation, (2) making energy efficiency a priority at the executive level of organizations will be largely ineffective if the IT and facilities staff are not directly incentivized to increase EE, and (3) there is minimal current understanding of how the individual psychologies of IT and facilities staff affect EE implementation in DCs. In the next phase of our research, we plan to interview data center operators/experts to ground-truth our literature findings and collaboratively design decarbonization policy solutions that target organizational structure, empower individual staff, and foster a supportive external market.
Chapter
Platforms for cloud computing make it possible for end users to get access to computing resources online. The emergence of cloud computing has led to an increase in the use of Apache Hadoop in cloud data centres. MapReduce and the open-source framework Hadoop have generated great interest among both researchers and corporates. Optimising Hadoop’s energy consumption has been identified as a critical issue, with numerous studies undertaken on the topic. FIFO, capacity, and fair scheduling algorithms are the most commonly used scheduling algorithms in Hadoop cloud data centres. This paper examines the nature and non-nature-oriented energy-efficient scheduling techniques that are frequently used in cloud data centres. The studies that have been explored are classified into node management, virtualisation and consolidation, task scheduling, data placement, and infrastructure change categories. A brief discussion of the rationale for each category is included, along with its strengths and weaknesses. Additionally, for computation and I/O-intensive workloads, the paper examines the CPU energy consumption of three commonly used cloud workload scheduling algorithms: FIFO, fair, and capacity. Our goal is to see if the energy usage among the three schedulers differs significantly between two types of workloads and different job sizes. Finally, the cloud computing research community has been provided with a futuristic perspective in the area, based on current trends in energy-efficient algorithms. The findings and directions presented in this study may be helpful to researchers interested in optimising cloud data centre energy consumption.
Preprint
Full-text available
Virtual machines are deployed to ease the performance, management overhead and access regularities of applications on cloud platforms. Virtual machines are often susceptible to overloading burdens, delays and other hurdles during server consolidation and migration processes. In order to regulate the energy dissipation, monitor the overloading and under loading issues, dynamic consolidation processes are introduced to distribute loads across virtual machines substantially. Consolidation process demands additional computations and resources to reallocate the services from one virtual machine to another, subjected to adherence of Service Level Agreements. The proposed methodology advocates the implementation of a novel architecture for consolidating virtual machines and thus balance the energy and performance parameters. Overall resource requirements, Performance to Power Ratio (PPR) are primary factors in design of proposed Dynamic Weightage algorithm, including the clustering approach with respect to reinforcement learning techniques. Cluster of optimal virtual machines are formed, resources are allocated based on optimal energy and performance expectations. Resource requests from virtual machines are derived into a matching relationship factor to represent the respective hosts with PPR considerations. These estimations also deliver the overall workload incurred during virtual machine consolidation. It is observed that the performance throughout the cluster is maintained to be a nominal with minimal trade-off to energy. The architecture is implemented over an offline platform, which are distributed environments, enabling scalability and improved efficiency of the entire system. The system is validated on CloudSim simulator with datasets retrieved from PlanetLab. From the results obtained, it is noted that energy conservation has yielded up to 47%, with promising quality of service parameters. Results are compared with other state of art algorithms for distributed architectures and heterogeneous environments to prove the efficiency of the system.
Article
Global warming and increased living standards have rapidly increased the demand for cooling systems. Meeting this demand in less economically developed areas is highly challenging due to a lack of electricity. In this paper, we have demonstrated a high-performance and wide relative humidity (RH) solar-driven evaporative cooling strategy that uses only atmospheric water. Developed here for the first time, we designed a metal–organic framework (MOF-801) based composite with a high-performance atmospheric water absorption across a wide range of RH. The as-synthesized composite can adsorb atmospheric water up to ~22% (~80%) of its weight at an RH@28% (~70%). Our demonstration has shown that the corresponding cooling powers range from 136 to 344 W/m² in a wide range of RH and solar intensities, and the passive cooling temperature is up to 14°C lower than the device's reference counterpart. Our study thus proposes a solar-driven cooling coating with high cooling powers across a wide range of RH based on an as-synthesized composite, which pinpoints a pathway to replace traditional compression-based cooling systems (e.g., air conditioners) and which will also have a significant impact in future global energy consumption.
Article
Full-text available
Background. There has been sustained and growing interest in characterizing the net energy impact of information and communication technology (ICT), which results from indirect effects offsetting (or amplifying) the energy directly consumed by ICT equipment. These indirect effects may be either positive or negative, and there is considerable disagreement as to the direction of this sign as well as the effect magnitude. Literature in this area ranges from studies focused on a single service (such as e-commerce versus traditional retail) to macroeconomic studies attempting to characterize the overall impact of ICT. Methods. We review the literature on the indirect energy effect of ICT found via Google Scholar, our own research, and input from other researchers in the field. The various studies are linked to an effect taxonomy, which is synthesized from several different hierarchies present in the literature. References are further grouped according to ICT service (e.g., e-commerce, telework) and summarized by scope, method, and quantitative and qualitative findings. Review results. Uncertainty persists in understanding the net energy effects of ICT. Results of indirect energy effect studies are highly sensitive to scoping decisions and assumptions made by the analyst. Uncertainty increases as the impact scope broadens, due to complex and interconnected effects. However, there is general agreement that ICT has large energy savings potential, but that the realization of this potential is highly dependent on deployment details and user behavior. Discussion. While the overall net effect of ICT is likely to remain unknown, this review suggests several guidelines for improving research quality in this area, including increased data collection, enhancing traditional modeling studies with sensitivity analysis, greater care in scoping, less confidence in characterizing aggregate impacts, more effort on understanding user behavior, and more contextual integration across the different levels of the effect taxonomy.
Article
Full-text available
Energy efficiency policies are pursued as a way to provide affordable and sustainable energy services. Efficiency measures that reduce energy service costs will free up resources that can be spent in the form of increased consumption - either of that same good or service or of other goods and services that require energy (and that have associated emissions). This is called the rebound effect. There is still significant ambiguity about how the rebound effect should be defined, how we can measure it, and how we can characterize its uncertainty. Occasionally the debate regarding its importance reemerges, in part because the existing studies are not easily comparable. The scope, region, end-uses, time period of analysis, and drivers for efficiency improvements all differ widely from study to study. As a result, listing one single number for rebound effects would be misleading. Rebound effects are likely to depend on the specific attributes of the policies that trigger the efficiency improvement, but such factors are often ignored. Implications for welfare changes resulting from rebound have also been largely ignored in the literature until recently.
Book
Full-text available
ICT Innovations for Sustainability is an investigation of how information and communication technology can contribute to sustainable development. It presents clear definitions of sustainability, suggesting conceptual frameworks for the positive and negative effects of ICT on sustainable development. It reviews methods of assessing the direct and indirect impact of ICT systems on energy and materials demand, and examines the results of such assessments. In addition, it investigates ICT-based approaches to supporting sustainable patterns of production and consumption, analyzing them at various levels of abstraction – from end-user devices, Internet infrastructure, user behavior, and social practices to macro-economic indicators. Combining approaches from Computer Science, Information Systems, Human-Computer Interaction, Economics, and Environmental Sciences, the book presents a new, holistic perspective on ICT for Sustainability (ICT4S). It is an indispensable resource for anyone working in the area of ICT for Energy Efficiency, Life Cycle Assessment of ICT, Green IT, Green Information Systems, Environmental Informatics, Energy Informatics, Sustainable HCI, or Computational Sustainability.
Article
Full-text available
Until recently, the main environmental concerns associated with information and communication technologies (ICTs) have been their use-phase electricity consumption and the chemicals associated with their manufacture, and the environmental effects of these technologies on other parts of the economy have largely been ignored. With the advent of mobile computing, communication, and sensing devices, these indirect effects have the potential to be much more important than the impacts from the use and manufacturing phases of this equipment. This article summarizes the trends that have propelled modern technological societies into the ultralow-power design space and explores the implications of these trends for the direct and indirect environmental impacts associated with these new technologies. It reviews the literature on environmental effects of information technology (also with an emphasis on low-power systems) and suggests areas for further research.
Article
The power usage effectiveness (PUE) metric has become an industry standard for reporting energy performance of data centers. However, it is an incomplete metric, failing to address hardware efficiency, energy productivity, and environmental performance. The industry should focus on adopting and systematically reporting more comprehensive metrics, which would allow more insight into data center energy performance.
Article
User-facing, latency-sensitive services, such as websearch, underutilize their computing resources during daily periods of low traffic. Reusing those resources for other tasks is rarely done in production services since the contention for shared resources can cause latency spikes that violate the service-level objectives of latency-sensitive tasks. The resulting under-utilization hurts both the affordability and energy efficiency of large-scale datacenters. With the slowdown in technology scaling caused by the sunsetting of Moore’s law, it becomes important to address this opportunity. We present Heracles, a feedback-based controller that enables the safe colocation of best-effort tasks alongside a latency-critical service. Heracles dynamically manages multiple hardware and software isolation mechanisms, such as CPU, memory, and network isolation, to ensure that the latency-sensitive job meets latency targets while maximizing the resources given to best-effort tasks. We evaluate Heracles using production latency-critical and batch workloads from Google and demonstrate average server utilizations of 90&percnt; without latency violations across all the load and colocation scenarios that we evaluated.
Chapter
Green communications has embraced data centers fully, and data center operators are seeking the most effective means to minimize the energy consumption of their data center infrastructure without impacting the performance of deployed digital services. In this chapter we discuss recent techniques that have the potential to improve the energy efficiency of data center networks. We then propose a network-centric energy management solution that combines guided routing to consolidate network flows on a compact set of network elements with fine-granular sleep mode techniques, working jointly with server power management and without affecting application performance. We finally discuss measurements obtained from a simulation of an OpenFlow-based implementation showing that our approach achieves significant energy savings under typical data center application loads, especially when both server and network power management are used together.