Assessing Internet energy intensity: A review of methods and results

https://doi.org/10.1016/j.eiar.2013.12.004 Get rights and content

Highlights

  • Assessments of the energy intensity of the Internet differ by a factor of 20,000.

  • We review top–down, model-based, and bottom–up estimates from literature.

  • Main divergence factors are the year studied and the inclusion of end devices.

  • We argue against extending the Internet system boundary beyond data transmission.

  • Decision-makers need data that differentiates between end devices and transmission.

Abstract

Assessing the average energy intensity of Internet transmissions is a complex task that has been a controversial subject of discussion. Estimates published over the last decade diverge by up to four orders of magnitude — from 0.0064 kilowatt-hours per gigabyte (kWh/GB) to 136 kWh/GB. This article presents a review of the methodological approaches used so far in such assessments: i) top–down analyses based on estimates of the overall Internet energy consumption and the overall Internet traffic, whereby average energy intensity is calculated by dividing energy by traffic for a given period of time, ii) model-based approaches that model all components needed to sustain an amount of Internet traffic, and iii) bottom–up approaches based on case studies and generalization of the results. Our analysis of the existing studies shows that the large spread of results is mainly caused by two factors: a) the year of reference of the analysis, which has significant influence due to efficiency gains in electronic equipment, and b) whether end devices such as personal computers or servers are included within the system boundary or not. For an overall assessment of the energy needed to perform a specific task involving the Internet, it is necessary to account for the types of end devices needed for the task, while the energy needed for data transmission can be added based on a generic estimate of Internet energy intensity for a given year. Separating the Internet as a data transmission system from the end devices leads to more accurate models and to results that are more informative for decision makers, because end devices and the networking equipment of the Internet usually belong to different spheres of control.

Introduction

Information and communication technologies (ICT) are credited as potentially important contributors towards a low-carbon economy (Erdmann and Hilty, 2010, European Commission, 2008, Hilty et al., 2013, Laitner, 2010, Laitner and Ehrhardt-Martinez, 2009, Mattern et al., 2010, Pamlin and Pahlman, 2008). There are, however, important controversies regarding the energy consumption and the environmental impact of some parts of ICT themselves. The energy intensity of the Internet, expressed as energy consumed per data transmitted, is one of these controversial issues.

The Internet is the infrastructure that connects billions of computers worldwide using the TCP/IP family of communication protocols. Some studies we are reviewing include the end devices communicating through the Internet (e.g., PCs and web servers) and their energy consumption in their object of research and therefore answer a different question than the studies addressing only the energy used for Internet data transmission.

Before we go into detail, we would like to point out that we are comparing estimates for direct energy consumption in the form of electricity only. The energy supply chain, containing the supply of primary energy, power plants transforming it to electricity, and grids bringing them to the consuming devices, is excluded from the system under study —although the electricity mix used, e.g., by data centers, is an issue of rising importance. We also exclude the “gray” energy embedded in ICT hardware, although the material flows caused by producing and disposing of hardware are significant (Hilty et al., 2011, Schluep et al., 2013). Furthermore, this study does not account for the increasing role of mobile Internet access and its impact on energy intensity. This restricted scope represents the least common denominator of the studies we analyzed.

Existing studies of Internet energy intensity (energy consumption per data transferred) lead to estimates differing by more than four orders of magnitude — from 136 kWh/GB (Koomey et al., 2004) down to 0.0064 kWh/GB (Baliga et al., 2011), a factor of more than 20,000.

This article presents a review of the existing studies, provides explanations for their large spread and concludes with a recommendation on how the system boundary for such studies should be drawn to be most useful for decision-making.

Section snippets

Assessments of Internet energy intensity

Our review includes the following ten studies published between 2004 and 2013: Baliga et al. (2007), Baliga et al. (2009), Baliga et al. (2011), Coroama et al. (2013), Koomey et al. (2004), Lanzisera et al. (2012), Pickavet et al. (2008), Schien et al. (2012), Taylor and Koomey (2008), Weber et al. (2010). Not all of them address the same object of research because they define different system boundaries for “the Internet”, a crucial issue to which we will return later. Some recent studies,

Comparison of the studies and explanation of diverging results

The surveyed studies present a very large variation among their results: from the 136 kWh/GB of (Koomey et al., 2004) down to the 0.006 kWh/GB of (Baliga et al., 2011), there is a spread of four orders of magnitude. Table 1 summarizes the most important characteristics and the results of the studies surveyed in this review. Some of the energy intensity results had to be calculated by the current authors to make the studies compatible, as described in Section 2 above. These values are marked with

Discussion

Our analysis has shown that there is one outstanding problem that must be solved when assessing the energy intensity of the Internet: the definition of the system boundary. The extreme case to view all ICT equipment connected to the Internet as part of the Internet leads to results that include energy consumption needed for various types of devices and tasks. Average energy intensity, in this case, does not say much about a specific task carried out by specific devices because of the high

Conclusion and outlook

We reviewed ten studies that assessed the average energy intensity of the Internet or quantities from which we could derive such an estimate. We found that the reference year had an obvious influence on the result. The most important conclusion is that the decision to either include end devices into such an assessment or to define the Internet as a pure data transmission system is crucial, both for the order of magnitude of the results as well as for the usability of the result to assess

Acknowledgments

This research was partly funded by the “Fundação para a Ciencia e Tecnologia”, Portugal, through project Pest-OE/EEI/LA0009/2011.

Vlad C. Coroama is a research fellow at the Instituto Superior Técnico (IST), Universidade Técnica de Lisboa, Portugal. He holds a computer science MSc degree from the Technical University of Darmstadt, Germany, and a PhD from the ETH Zurich, Switzerland. For more than a decade, his research revolved around the relation between Information and Communication Technologies (ICT) and sustainability. In recent years, he focused exclusively on the environmental dimension of sustainability. Vlad is

References (44)

  • J.G. Bull et al.

    Comparative life cycle assessments: the case of paper and digital media

    Environ Impact Assess Rev

    (2014)
  • V.C. Coroama et al.

    Effects of Internet-based multiple-site conferences on greenhouse gas emissions

    Telematics Inform

    (2012)
  • G. Finnveden et al.

    Recent developments in life cycle assessment

    J Environ Manag

    (2009)
  • Accenture

    Cloud computing and sustainability: the environmental benefits of moving to the cloud

    (2010)
  • J. Baliga et al.

    Energy consumption of the Internet, Joint International Conferences on Optical Internet, and the 32nd Australian Conference on Optical Fibre Technology, COIN-ACOFT

    (2007)
  • J. Baliga et al.

    Energy consumption in access networks. Optical Fiber Communication Conference and the National Fiber Optic Engineers Conference (OFC/NFOEC)

    (2008)
  • J. Baliga et al.

    Energy consumption in optical IP networks

    J Light Technol

    (2009)
  • J. Baliga et al.

    Green cloud computing: balancing energy in processing, storage, and transport

    Proc IEEE

    (2011)
  • CEET

    The power of wireless cloud: an analysis of the impact on energy consumption of the growing popularity of accessing cloud services via wireless devices

    (2013)
  • C.A. Chan et al.

    Methodologies for assessing the use-phase power consumption and greenhouse gas emissions of telecommunications network services

    Environ Sci Technol

    (2012)
  • Cisco

    Hyperconnectivity and the approaching zettabyte era

    (2010)
  • Cisco

    VNI forecast highlights: Australia — 2010 year in review

    (2011)
  • G. Cook

    How clean is your cloud?

    (2012)
  • V.C. Coroama et al.

    The direct energy demand of Internet data flows

    J Ind Ecol

    (2013)
  • L. Erdmann et al.

    Scenario analysis: exploring the macroeconomic impacts of information and communication technologies on greenhouse gas emissions

    J Ind Ecol

    (2010)
  • European Commission

    Addressing the challenge of energy efficiency through information and communication technologies

    (2008)
  • Google

    Google's green computing: efficiency at scale

    (2011)
  • Google

    Google apps: energy efficiency in the cloud

    (2012)
  • L. Hilty et al.

    Sustainability and ICT—an overview of the field

    Notizie Politeia

    (2011)
  • K. Hinton et al.

    Power consumption and energy efficiency in the Internet

    IEEE Netw

    (2011)
  • K. Hinton et al.

    Energy efficiency of optical IP protocol suites

    (2012)
  • Cited by (76)

    View all citing articles on Scopus

    Vlad C. Coroama is a research fellow at the Instituto Superior Técnico (IST), Universidade Técnica de Lisboa, Portugal. He holds a computer science MSc degree from the Technical University of Darmstadt, Germany, and a PhD from the ETH Zurich, Switzerland. For more than a decade, his research revolved around the relation between Information and Communication Technologies (ICT) and sustainability. In recent years, he focused exclusively on the environmental dimension of sustainability. Vlad is the author of several studies and articles on the environmental impact assessment of ICT in general and the Internet in particular.

    Lorenz M. Hilty is professor at the Department of Informatics at the University of Zurich and senior scientist at Empa, the Swiss Federal Laboratories for Materials Science and Technology. His interdisciplinary team, the Informatics and Sustainability Research (ISR) group, is shared by the University of Zurich and Empa. He is affiliate professor at the Center for Sustainable Communications (CESC) at KTH Royal Institute of Technology, Stockholm. Lorenz Hilty got his Dr. rer. nat. and habilitation in Computer Science from the University of Hamburg. He has published more than 100 papers and books in the field of ICT and sustainability.

    View full text