Scope
DIMPACT is a private sector initiative collaborative project convened by Carnstone and the University of Bristol to measure and report the carbon footprint of digital services. Some DIMPACT participants are considered part of the ICT sector – or have segments of their business that fall in this category – whereas others are wholly part of the Entertainment and Media sectors. DIMPACT participants include some of the largest media companies in the world, including Netflix, the BBC and the Economist. Currently, the DIMPACT tool estimates the emissions associated with video streaming, online banner advertising, digital publishing, and audio streaming, but they are also working to include capabilities to measure audio streaming and video conferencing soon.
DIMPACT reports [1, 2], introduce several energy consumption and carbon footprint aspects of Information and Communication technology (ICT) related to EXIGENCE interests and scope, including the following.
Summary
Energy consumption of data centres and networks currently represents about 2-3% of global electricity consumption, driving 0.6% of total greenhouse gas (GHG) emissions, under the specific assumptions of DIMPACT reports. We must still strive for efficiency everywhere (especially from screens and personal devices), but increased demand for digital services has not caused proportional growth in energy consumption or carbon emissions. Due to modelling and data access limitations, there are complexities in the relationship between data traffic, energy consumption, and carbon emissions that require careful consideration before making real-world decisions. Actions towards decarbonisation should, therefore, be driven by stakeholder-specific evidence and state-of-the-art data to avoid potential unintended adverse consequences.
DIMPACT Draft Paper “Literature Review and Policy Principles for Streaming and Digital Media Carbon Footprinting” (March 2023) [1] outlines the policy suggestions based on DIMPACT’s experience and review of the literature, as well as, the latest technical and methodological thinking on digital emissions, upon which the policy recommendations are based.
The following guiding principles for interested stakeholders and policymakers are proposed.
Principle 1: Expand access to shared, contemporary data
- Enable standardised data sharing across the digital sector through data reporting and aggregation protocols. This will enable better holistic decision-making and ensure the use of relevant data that best represents the current state of the rapidly evolving digital sector. Data and methodologies should be standardised across operators so that metrics can be aggregated to understand the impacts across the sector without compromising competitive information. Data should be collected in a way that allows for better demand forecasting.
- Leverage contemporary data (less than 1-2 years old) to help inform future decisions in a rapidly changing sector.
Principle 2: Ensure appropriate modelling for short- and long-term decision making
- Conduct additional research on demand response (peak vs off-peak internet use), i.e. the time-variable throughput of data traffic through the digital sector and network, and account for the network’s baseload of energy (for mobile and fixed networks).
- Use appropriate modelling for changes to digital sector energy use that reflects the energy dynamics of these systems. Models, such as the Power Model, can assess and allocate networks’ energy to a given service, reflecting the observed immediate response between data transmission rates and energy.
Principle 3: Institute energy efficiency incentives for devices and infrastructure
- Promote energy-efficient devices and infrastructure, including TVs, data servers, data centre cooling, networks, and in-home devices, especially for devices powered on all the time, even when idle. Media and entertainment companies can play an indirect role by influencing their value chain partners to adopt more efficient technology – and, to a smaller degree, directly influence the energy consumption of specific user devices. Increasing the efficiency of screen devices will significantly reduce the end-to-end energy consumption of streaming and other digital activities.
- Incentivise efficient device and infrastructure utilisation. Generally, the fewer connected devices we use, the more efficiently the digital sector operates. For example, network technology that reduces the need for personal infrastructure (e.g., Wi-Fi routers in every home) may reduce overall energy consumption.
Principle 4: Prioritise broad availability of Low Carbon and Renewable Energy
- Invest in low-carbon and renewable energy infrastructure so that corporations can set and achieve sustainability goals and accelerate the transition to low-carbon electricity. Renewable electricity could reduce ICT emissions by 80%.
- Enable low-cost renewable energy for at-home usage; this will have the most impact on video streaming.
Opportunities to reduce the digital sector’s use-phase carbon emissions:
- Energy Efficiency across all energy-consuming devices in the value chain, including data centres, network infrastructure, and home devices such as Wifi routers, TVs, laptops and peripherals.
- Holistically optimising digital content delivery. This includes consolidating data centres and content distribution networks into efficient facilities with efficient hardware and increasing network infrastructure’s energy efficiency and utilisation.
- Low-carbon energy implementation is available wherever possible, especially for the individual consumer and the screen they’re using.
So far, two typical allocation methodologies have been applied to allocate the energy consumption and emissions of data transmission networks: Allocating based on average data volumes (the Average Data Method) and allocating based on time (the Power Model).
The Power Model
The Power Model [3] gives a better sense of the short-term marginal change in network energy consumption based on changes in data volume transfers (e.g. switching a video stream from 4K to SD). This Power Model acknowledges that there is a high baseload energy required to keep ICT networks running for all its users, whether or not data is flowing through it, based on time duration of usage and the number of subscription lines (estimated, for fixed line networks, at +0.02W/line when browsing the web for lower data volumes, vs. +0.2W/line when streaming Netflix at 4 Mbps). The model does not take into account peak data volumes. This model cannot predict long-term trends.
Average Data Attributional Method (the DIMPACT model [2])
The Average Data Method can only be used to determine how much of the total ICT sector’s energy should be attributed to a single activity that happened in the past. A model for attributing historical, system-level carbon emissions to video streaming based on (a) streaming-related data centre energy consumption, (b) streaming-related network data traffic, and (c) power consumption from screens and connected devices. A key finding is that end-user devices are the primary driver of use-phase emissions under the specific assumptions of DIMPACT reports. Data centres contribute <1% and networks contribute ~10% to emissions, which means that the remainder comes from energy consumed by user devices, including TVs, internet modems and Wi-Fi routers. This model cannot be used to determine the real-world consequence of a change, e.g., varying data volume levels. It does not consider that approximately 80% of total electricity in fixed networks comes from base load.
These are the main equations for the average data attributional method>
- Equation 1. Use-phase GHG emissions of data centre components:
Data center component use-phase GHG emissions = electrical energy per period (kWh/period) * carbon intensity electricity used (kgCO₂e/kWh)
- Equation 2. Calculation of energy consumption of internet infrastructure:
Internet energy consumption = data volume transferred (GB) * intensity (kWh/GB)
- Equation 3. Estimating data volume transferred for webpages and mobile apps:
Data volume transferred = mean data volume per pageview * number of pageviews per period
Data volume transferred = mean data volume per app session * number of app sessions per period
- Equation 4. Estimating data volume transferred for downloads:
Data volume transferred = mean data volume per download * number of downloads per period
- Equation 5. Estimating data volume transferred for streamed video:
Data volume transferred = mean bitrate of content served * duration of content served per period
Internet energy consumption is estimated by DIMPACT using the approach described in the Greenhouse Gas Protocol ICT Sector Guidance (Chapter 4, p26). This approach uses a single intensity metric based on the data transferred in kWh/GB, with different values for fixed-line and cellular (mobile) networks.
“The internet emissions may be calculated by using an energy intensity factor for the internet (expressed in kWh/GB) and multiplying this by the data transferred (in GB) and an electricity emission factor (in kg CO₂e / kWh).”
- Equation 6. Estimating the carbon emissions of internet energy consumption:
Carbon Footprint (kgCO₂e/period) = Carbon intensity of electricity (kgCO₂e/kWh) * Electrical energy used in data transfer in period (kWh/period)
- Equation 7. Estimating energy consumption allocation of Content delivery networks (CDNs):
Electrical energy used in CDN services in period (kWh/period)= Electrical Energy intensity of CDN Services (kWh/GB) * Data volume served by CDN service per period (GB per period)
- Equation 8. Estimating the GHG emissions of CDNs:
GHG emissions (kgCO₂e/period) = Carbon intensity of electricity (kgCO₂e/kWh) * Electrical energy used in CDN services in period (kWh/period)
- Equation 9. Estimating the total energy of Customer premises equipment (CPE):
Total mean energy of CPE per period per premises (kWh/period) = Mean power of CPE on premises (kW) * Duration of period (hours per period)
- Equation 10. Estimating energy intensity for CPE:
Mean Electrical Energy used by CPE per GB (kWh/GB) = Total mean energy of CPE per period per premises (kWh/period) / Total mean data volume used per premises per period (GB/period)
- Equation 11. Allocating energy consumption for CPE:
Estimated CPE electrical energy allocated to service (kWh/period) = Mean Electrical Energy used by CPE per GB (kWh/GB) * Data volume transferred over fixed-line networks per period (GB/period)
- Equation 12. Estimating GHG emissions for CPE:
Carbon Footprint (kgCO₂e/period) = Carbon intensity of electricity (kgCO₂e/kWh) * Estimated CPE electrical energy allocated to service (kWh/period)
- Equation 13. Energy consumption of end-user devices:
Device energy consumption (kWh) = Duration of service (W) x Power consumption of device (hrs) / 1000 (W/kWh)
- Equation 14. GHG emissions of end-user devices:
GHG emissions (kgCO₂e) = Energy consumption (kWh)* GHG emissions factor for electricity used in users’ homes (kgCO₂e/kWh)
- Equation 15. Estimating service duration of end-user devices:
Mean service duration per period on device type = Number of views/sessions per period (no) * Mean duration per view/session
Or: Mean service duration per period on device type = duration as measured by user analytics systems directly
- Equation 16. Estimating end-user device energy:
End-user device energy for device type = Mean power of device type (W) * mean service duration per period on device type (hrs per period)
- Equation 17. Estimating GHG emissions from end-user devices:
End-user device carbon footprint for device type = End-user device energy for device type * Carbon intensity of electricity (kgCO₂e/kWh)
- Equation 18. Estimating overall standby power of end-user devices:
Total mean standby energy of device per period (kWh/period) = Mean power device in standby mode (kW) * Duration of device in standby mode per period (hours per period)
- Equation 19. Allocating standby power to a given service:
Allocated standby energy to service per period (kWh/period) = Total mean standby energy of device per period (kWh/period) * [Total duration of use by service per period / Total duration of use for all services per period]
- Equation 20. Estimating the GHG emissions of allocated standby power:
Allocated standby carbon footprint (kgCO₂e/period) = Allocated standby energy to service per period (kWh/period) * Carbon intensity of electricity (kgCO₂e/kWh).
Areas of Future Investigation:
- Peak Demand [1]: What drives peak demand, and how different stakeholders in the value chain can collaborate to encourage demand shifting, i.e. better utilising networks to prevent peak events and capacity constraints.
- How we can more accurately model the longer-term energy and carbon impacts of a peak event causing network operators to build capacity in their networks.
- Schien et al. [4] propose an alternative methodology for establishing an energy intensity metric that “redistributes [the] burden of baseline power consumption proportional to data throughput” by reallocating the baseload’s energy to periods of peak demand, thereby accounting for the influence of bandwidth demand on real-world scenarios.
- Network Type [1]: All digital sector energy and carbon emission models today are limited by a lack of data on the influence of network type. For fixed networks, we know that energy loads today are generally unaffected by system-wide data traffic. In contrast, it is currently very difficult for mobile networks to attribute usage to individual users, and existing studies today show mixed results on the pros and cons of mobile versus fixed networks. Network providers should provide up-to-date data that will help address the uncertainties in current modelling. These include:
- Power per subscriber (Watts per subscriber line) for each connection type (e.g. FTTH, xDSL) in a standardised format.
- Power consumption of networks during peak-, average- and low-traffic scenarios.
- Equivalent usage metrics (total data volume per household, number of subscribers per connection type, peak capacity).
- Non-human end-users [2]: bots, crawlers and spiders. This presents a challenge because of the myriad behaviours of these bots in terms of the types of requests made and the amount of content gathered (e.g., full webpage versus headers or even simply receiving an error message). Understanding the scale of these bots requires further investigation.
References
[1] DIMPACT Draft Paper “Literature Review and Policy Principles for Streaming and Digital Media Carbon Footprinting” (March 2023).
[2] DIMPACT Draft Paper “Methodology: Estimating the carbon impacts of serving digital media and entertainment products” (October 2022)
[3] Malmodin, J. (2020, September). The power consumption of mobile and fixed network data services-The case of streaming video and downloading large files. In Electronics Goes Green (Vol. 2020).
[4] Schien, Preist & Shabajee (2022), “Rethinking Allocation in High-Baseload Systems: A Demand-Proportional Network Electricity Intensity Metric”, Position Paper for IAB workshop on Environmental Impact of Internet Applications and Systems.
(October 2022).
Relevance for EXIGENCE
Energy efficiency and carbon footprint of related ICT aspects are relevant for requirements and scenarios, energy metrics, energy measurements, orchestration and energy optimisation.