↓Skip to main content

Paper Notes – The overlooked environmental footprint of increasing Internet use

·
Paper Notes Cloud IT Energy Data Center Energy Network Energy
Table of Contents

Paper #

Obringer, R., Rachunok, B., Maia-Silva, D., Arbabzadeh, M., Nateghi, R., and Madani, K. (2021). The overlooked environmental footprint of increasing Internet use. Resour. Conserv. Recycl. 167, 105389. https://doi.org/10.1016/j.resconrec.2020.105389

Notes #

  • The article itself has no abstract and is very short. I assume this is because it is a “Perspective” article, so most of what is worth looking at is contained within the supplementary materials.
  • The article makes an estimate of the carbon, water and land footprints of fixed-line internet use. It estimates that “Globally, the Internet use has a carbon footprint ranging from 28 to 63 g CO2 equivalent per gigabyte (GB), while its water and land foot-prints range from 0.1 to 35 L/GB and 0.7 to 20 cm2/GB, respectively”.
  • The system boundary only includes fixed-line networks, thereby excluding all end-user devices as well as mobile connections. This is a valid system boundary, but excluding mobile devices misses a major source of network traffic.
  • The value for energy intensity of the network traffic is taken from Aslan, 2018. This matches the system boundary (Aslan only considers fixed-line networking). However, they do not make any adjustments to the energy efficiency. Aslan clearly states that the trend is for network energy to fall by 50% every 2 years, but that is not taken into account.
  • A reference is also made to “The carbon footprint of streaming video: fact-checking the headlines” from the IEA, but the network energy intensity values do not appear to be taken from that article. The IEA give a figure of 0.002 kWh/GB, but Obringer et al use 0.06 kWh/GB (which is the value for 2015 from Aslan).
  • They include data centers as a component of the calculation, but use network traffic as a proxy for the energy consumption at an intensity of 0.01 kWh/GB. I couldn’t figure out where this number comes from. This results in a per GB energy intensity figure, including the data center, rather than a total figure of annual data center (or network) energy consumption.
  • Calculations are made based on an assumed linear correlation between network traffic and energy consumption. Energy intensity factors are multiplied by hourly estimated data transmission volumes for a range of applications. This includes the data center component, which doesn’t make much sense – data volume is rarely related to the energy consumption of a data center – servers, storage, etc because there are too many other processes involved. Transmission of 1GB of data could result in optimization, rendering, machine learning training and inference, redundant storage and replication, extraction of just a single value from that data…and so on.
  • However, they also state:

In practice, energy use for data storage and transmission does not increase linearly with increased internet use. Our estimation disregards this nonlinearity and relies on average energy use per gigabyte of data… Similarly, the footprints of energy production do not change linearly with increased energy use. Our estimations disregard this nonlinearity as well as other factors, such as the change in the energy mix by the change in level of production or hour of use (e.g., during peak hours).

  • Even though they reference three other publications which they claim to take the same approach, this assumption invalidates their whole methodology. It seems very strange that you would acknowledge that there is no linear correlation between network energy and data transmission, then disregard that and proceed with an approach on the basis that there is a correlation. The IEA source they cite to back up this approach specifically says “the latest research shows that these data-based intensity values (kWh/GB) are not appropriate for estimating the network energy use of high bitrate applications such as streaming video”.
  • At the time of writing, the IEA article was referring to informal sources to make that statement. These have now been published more formally, such as in Malmodin, 2020 and Koomey & Masanet, 2021.
  • There is an argument to be made that there will be an increase in energy consumption as network traffic increases, just it is not likely to be linear. Network operators provision for peak load, and redundancy, so growth in traffic will eventually result in new network equipment being deployed to cope with that increase. This may result in a decrease in energy intensity (newer equipment can handle more traffic), but it may need more equipment to be deployed (so an increase in the absolute energy consumption).
  • Obringer et al. do not make that argument.

Conclusions #

Unfortunately the methodological flaws mean the results aren’t useful.

↑