Energy consumption of network communication

Blog Article

Together with Arne Tarara I wrote a blog article about this topic:
www.green-coding.io/blog/network-carbon-emissions-in-green-software/

System boundaries

The system boundaries are set differently depending on the study.

Examples of different system boundaries:

See also the network categories defined by the Technology Carbon Standard:

Energy consumption of internet data transfer

Data transfer != network energy usage

Often the metric kWh/GB is used to calculate the energy consumption of network transfer. However, there is no correlation between the amount of transferred data and energy consumption! Therefore it can't be used for consequential reasoning (halve the data transfer != halve the emissions).

Energy consumption per transferred data is a bad metric

Coroama 2021

coroama.2021.investigatinginconsistenciesenergy (pg. 9)

The energy intensity of the Internet (EI) has been addressed in several studies, such as (Schien et al. 2013; Coroama et al. 2013). For fixed access networks, the current academic state-of-the-art indicates an average energy intensity of 0.007 kWh/GB for 2020 (Aslan et al. 2018). For the energy consumption of networks as a whole, on the other hand, there is some divergence in the academic literature, ranging from 325 TWh yearly (Malmodin and Lundén 2018) to 723 TWh (Andrae and Edler 2015).

coroama.2021.investigatinginconsistenciesenergy (image) (pg. 14)

Figure 3 High-level topology of the Internet, distinguishing between data centres (DCs), end devices, and three distinct types of the  network: the wide-area network (WAN), fixed access network (FAN), and radio access network (RAN).

coroama.2021.investigatinginconsistenciesenergy (pg. 20)

As stated in the Introduction, the study by (Aslan et al. 2018) is the current state of the art of the academic understanding on the energy intensity of the WAN. Extrapolating exponential past trends (up to 2017) via least squares fit yields a value of around EI2019 (WAN) = 0.01 kWh/GB and EI2020 (WAN) = 0.007 kWh/GB. The study does not address the energy intensities of either FAN or RAN.

coroama.2021.investigatinginconsistenciesenergy (pg. 20)

(Pihkola et al. 2018) also extrapolate their 2010-2017 results into the future, devising an energy intensity of just below EI2020 (RAN) = 0.1 kWh/GB.

coroama.2021.investigatinginconsistenciesenergy (image) (pg. 24)

Table 1 Summary of existing estimates for EI (WAN), EI (FAN), and EI (RAN) for the last decade (2010 – 2019) and for today (2021).

coroama.2021.investigatinginconsistenciesenergy (pg. 37)

These estimates, whose exact values are presented in row 10 of Table 3 and in Section 5.2, can be rounded as follows:

  • EI2020 (WAN) = 0.02 kWh/GB,
  • EI2020 (FAN) = 0.07 kWh/GB,
  • EI2020 (RAN) = 0.2 kWh/GB,
  • E2020 (WAN) = 110 TWh,
  • E2020 (FAN) = 130 TWh,
  • E2020 (RAN) = 100 TWh
coroama.2021.investigatinginconsistenciesenergy (pg. 38)

These numbers probably also represent a reasonable basis for extrapolation for a couple of years into the future. The yearly efficiency gains (corresponding to the yearly decreases of the energy intensity) need to be taken into account according to the rather trivial equation:

(14)
EIᵧ₊₁(X) = eirfᵧ(X) × EIᵧ(X)

for all X ∈ {WAN, FAN, RAN}, and where eirfᵧ(X) represents the yearly energy intensity reduction factor for the year y.

There are approximations of the intensity reduction factors in the literature; their caveat, however, is that – as shown above – they typically resulted in too optimistic results in the past. Until updated factors validated against top-down studies emerge, we suggest using slightly more pessimistic values than those of the literature, as follows:

  1. eirf₂₀₂₀(WAN) = 0.7 (Aslan et al. 2018) + 0.1 = 0.8
  2. eirf₂₀₂₀(FAN) = 0.75 (Wu 2021) + 0.1 = 0.85
  3. eirf₂₀₂₀(RAN) = 0.7 (Pihkola et al. 2018) − 0.8 (Elisa 2020)
    Choose the more pessimistic and directly measured value: 0.8.

Of course, Equation (14) can be used for several years if assuming a constant reduction factor:

(15)
EIᵧ₊ₙ(X) = (eirfᵧ(X))ⁿ × EIᵧ(X)

This is a reasonable assumption for a couple of years; equation (15) should not be used for a longer time horizon anyway, given the uncertainties of the current data (see also next section) and the need for mutual validation between energy and energy intensity values in the new studies to come, which will thus hopefully decrease current uncertainties.

Extrapolated numbers

Energy Intensity Factor:

WAN FAN RAN
Reduction Factor 0.8 0.85 0.8
2020 0.02 0.07 0.2 kWh/GB
2021 0.016 0.0595 0.16 kWh/GB
2022 0.0128 0.050575 0.128 kWh/GB
2023 0.01024 0.04298875 0.1024 kWh/GB
2024 0.008192 0.036540438 0.08192 kWh/GB
2025 0.0065536 0.031059372 0.065536 kWh/GB
2026 0.0052429 0.026400466 0.0524288 kWh/GB
2027 0.0041943 0.022440396 0.041943 kWh/GB
2028 0.0033554 0.019074337 0.0335544 kWh/GB
2029 0.0026844 0.016213186 0.0268435 kWh/GB

Network mix (example):

WAN FAN RAN Total
Network Mix 100% 90% 10%
2025 0.0065536 0.027953435 0.0065536 0.041060635 kWh/GB
2026 0.0052429 0.023760419 0.0052429 0.034246179 kWh/GB
2027 0.0041943 0.020196357 0.0041943 0.028584965 kWh/GB
2028 0.0033554 0.017166903 0.0033554 0.023877789 kWh/GB
2029 0.0026844 0.014591868 0.0026844 0.019960577 kWh/GB

Aslan et al. (2018)

Scope: Transmission network

aslan.etal.2018.electricityintensityinternet (comment) (pg. 3)

Transmission network = IP core network (e.g. core/metro/edge switches and routers) + access network (e.g. DSLAM)

Not included: undersea cables, data centers, home/on-site network equipment, user devices

Aslan et al. estimate that data transmission costs decrease by 50% every two years.
Quote:

aslan.etal.2018.electricityintensityinternet (pg. 13)

Estimates for average transmission network electricity intensity that meet these criteria show a halving of intensity every 2 years.

stephens.etal.2021.carbonimpactvideo (pg. 21)

Aslan’s rule
Analysis of estimates for the average electricity intensity of fixed-line internet transmission networks for data transfers from 2000 to 2015 concluded that electricity intensity (in kWh/GB) decreased by half approximately every two years over that time period (Aslan et al., 2018).

stephens.etal.2021.carbonimpactvideo (comment) (pg. 42)

Fixed network energy intensity (2020) = 0.0065 kWh/GB
Mobile network energy intensity (2020) = 0.1 kWh/GB
Home router energy intensity (2019) = 0.025 kWh/GB

Numbers based on Aslan's rule

Formulas:

Year kWh/GB
2015 0.06
2016 0.042426407
2017 0.03
2018 0.021213203
2019 0.015
2020 0.010606602
2021 0.0075
2022 0.005303301
2023 0.00375
2024 0.00265165
2025 0.001875
2026 0.001325825
2027 0.0009375
2028 0.000662913
2029 0.00046875

Andrae (2020)

Wireless access network: 0.18 kWh/GB
Fixed access wired networks: 0.07 kWh/GB

Energy consumption of mobile cellular communications

@Golard.etal.2023.EvaluationProjection4G5GRANEnergyFootprints

Projection for Belgium for 2020–2025 of radio access networks (RAN):

@Pihkola.etal.2018.EvaluatingEnergyConsumption

Mobile network in Finnland:

Energy efficiencies of different networks

Energy consumption of data transfer between data centers

Cloud Carbon Footprint

Networking

Source: https://www.cloudcarbonfootprint.org/docs/methodology/#networking (accessed on 2024-01-29)

Currently, our application takes into account only the data exchanged between different geographical data centers.
For networking, it is safe to assume that the electricity used to power the internal network is close to 0, or at least negligible compared to the electricity required to power servers.

[…]

There have not been many studies that deal specifically with estimating the electricity impact of exchanging data across data-centers. Most studies focus on estimating the impact of end-user traffic from the data center to the mobile phone; integrating the scope of the core network (what we are interested in), the local access to internet (optical fiber, copper, or 3G/4G/5G) and eventually the connection to the phone (WiFi or 4G).

[…]

It is safe to assume hyper-scale cloud providers have a very energy efficient network between their data centers with their own optical fiber networks and submarine cable. Data exchanges between data-centers are also done with a very high bitrate (~100 GbE -> 100 Gbps), thus being the most efficient use-case. Given these assumptions, we have decided to use the smallest coefficient available to date: 0.001 kWh/Gb.

Appendix IV: Recent Networking studies

Discussions:

Data exchange between data centers of the same hyper-scale cloud provider:
0.001 kWh/Gb

Embodied carbon on network infrastructure

Embodied Carbon of Network Infrastructure

Footprint of unwanted data transfer

Energy consumption of advertisements

Carbon footprint of unwanted data-use by smartphones (Uijttewaal et al.)

uijttewaal.etal.2021.carbonfootprintunwanted (pg. 12)

we estimate that the yearly carbon footprint of data-use by advertising and tracking services in smartphone apps in Europe is between 5 and 14 Mt CO2-eq.

Data used by website carbon calculators

The calculation model of SustainableWebDesign.org that is used by the CO2.js library and multiple website carbon calculators not only includes the energy used by the IP network, but also the energy used by consumer devices, data centers and the production of the hardware components.

See SustainableWebDesign.org#Emissions Calculation Formulas for more information.

Videostreaming

Climate emissions of videostreaming

What to do?

Romain Jacob:

Summary

  1. The Internet peak traffic increases. That is driving network growth, which results in emptier networks on average.
  2. Networking hardware is far from power proportionality; It is very inefficient at low utilization.
    → It can and must be improved.
  3. Even if direct cost of traffic is minuscule, sending more traffic has an important systemic cost. Cf. Point 1.
    → Don't send if you don't need to.
    → Stay away from the peak!