Table of Contents
My area of research interest could be broadly called “sustainable computing”, which I am actively pursuing in the Department of Engineering Science at the University of Oxford. I maintain this page as a set of resources about sustainable computing that I find useful.
What is sustainable computing? #
Sustainable computing concerns the consumption of computing resources in a way that means it has a net zero impact on the environment, a broad concept that includes energy, ecosystems, pollution and natural resources.
I consider that individual impact on the environment rounds to zero because although some people will change their behaviour, most won’t. This means that systems need to change. Energy production, transportation, manufacturing, agriculture. The transition to sustainability needs to be done on behalf of the individual by the system and at a large scale, so individuals can benefit from that change by default.
In computing, this means the organisations responsible for manufacturing components, building equipment, writing software, operating data centres, and running networks have responsibility for achieving sustainability.
Sustainability strategies in computing tend to focus on energy and the transition to clean energy – the use stage. This is important, but not the only factor.
Sustainable computing therefore involves not just clean energy, but also minimising water stress, developing policies around the right to repair and recycling of materials through a circular economy, energy efficiency of hardware and software, promoting good governance of resources, transparency, and consistent reporting.
- Estimating AWS EC2 Instances Power Consumption. This writeup explains a pragmatic approach to figuring out the energy consumption of AWS EC2 images. It uses measurements and some reverse engineering to build profiles for some common instance types, and reveals issues with relying on the SPECpower dataset (mentioned below).
- Building an AWS EC2 Carbon Emissions Dataset. A followup to the above post to add embodied emissions calculations to EC2 instances. In Dec 2021 I implemented this methodology into the Cloud Carbon Footprint project as part of the cloud carbon coefficients calculations.
- ClimateAction.tech. I subscribe to this weekly newsletter because it provides a good summary of interesting news, events and podcasts related to what the technology sector can do to help climate change. I’m not in the Slack group though because I try and avoid group chat where possible.
- Green Software Foundation. A group of technologists (primarily from commercial organizations) working on standards and best practices to help software developers build more sustainable software.
Interesting sources of data related to sustainable computing:
- Cambridge Bitcoin Electricity Consumption Index. The headline figure is 143 TWh of annualised energy consumption, but with a range of 65 TWh – 236 TWh. However the missing context is how that electricity is generated. Distinguishing between whether that came from fossil fuels or renewables is important (39% renewables, primarily hydro), particularly as governments impose stricter regulations which may have unintended negative effects (such as increasing carbon intensity of mining). Whether it is possible to move to a proof-of-stake implementation, like Ethereum, is also important.
- Carbon free energy for Google Cloud regions. Google is the leader in sustainable cloud computing. They have done the most work and made the most progress at reaching carbon neutrality, then 100% renewables matching, and have a goal to hit 24/7 carbon free by 2030. They are less transparent on water consumption (publishing some numbers for their data centers in the US), but do publish data on their carbon free energy percentage for each GCP region.
- SPECPower database. The Standard Performance Evaluation Corporation (SPEC) provides quarterly test results of the power and performance of the latest single and multi-node servers. By simulating workloads, this data shows the average power consumption (in watts) at 100% and idle load. The difference between the two is known as power proportionality – does the server reduce its power consumption in proportion to the load – a ratio which has been improving over time. This data is used as part of the cloud carbon coefficients project to provide estimates for the power consumption of popular cloud computing CPUs for use in the Cloud Carbon Footprint project.
Here are the most important academic papers related to sustainable computing, but I also publish notes about papers I read:
- ICT Sector Electricity Consumption and Greenhouse Gas Emissions – 2020 Outcome SSRN Electronic Journal (2023). This is the most up to date analysis of the total electricity consumption and GHG emissions of the whole ICT sector. It shows that ICT is estimated to account for ~4% of global electricity consumption in 2020. Data center energy consumption is estimated at 223 TWh and network energy consumption is estimated at 272 TWh.
- The environmental footprint of data centers in the United States. Environmental Research Letters. (2021). This paper discusses the wider impact of data centres but has some major water-related findings. They show that 1/5th of direct water footprint (for cooling) comes from moderate to high stress watersheds, and that nearly half of power generation is sourced from power plants in water stressed regions. This is an important analysis showing the impact of indirect water consumption through energy generation. See also: How much water do data centers use?
- Recalibrating global data center energy-use estimates. Science. (2020). This paper represents the current best figures of global data centre energy consumption from the perspective that energy growth has plateaued. They estimate data centres consumed 196 TWh of energy in 2020. Their system boundaries exclude cryptocurrencies and make some major assumptions about the migration to hyperscale cloud (AWS, Google, Microsoft).
- Green Cloud? Advances in Computer Science Research (2016). This paper represents the other end of the credible data centre energy estimates, placing data centre energy consumption at 287 TWh in 2015. Their system boundaries do include crypto. See also subsequent papers from the same authors: Energy consumption of data centers worldwide (2019) and Efficiency gains are not enough: Data center energy consumption continues to rise significantly (2020) with the last of these providing an estimate for 400 TWh for 2020. An upcoming publication mentioned in a brief research note may put the estimate at 350-500 TWh in 2021.
- Electricity Intensity of Internet Data Transmission: Untangling the Estimates. Journal of Industrial Ecology (2018). This is the latest calculation of fixed-line network energy intensity i.e. the energy consumption of data transmission. The figure for 2015 is 0.06 kWh/GB with a projection that this figure will fall by 50% every 2 years. The notable aspect of this system boundary is that it excludes mobile devices.
See also my own publications and Sources of data center energy estimates: A comprehensive review Joule (2022) in particular.
Interesting reports related to sustainable computing:
- 24/7 by 2030, Google (2020) This describes the challenges with procuring 100% carbon free energy and why 100% doesn’t actually mean 100%. It shows how this relates to data centers and the path towards achieving it. Google is leading the industry in this regard, but there are challenges for the energy system.
- Carbon impact of video streaming. 2021, Carbon Trust This is the most accurate analysis of the energy and carbon footprint of video streaming. It was written through an industry partnership with Netflix, so uses validated data, but based on academic models through work done at the University of Bristol. It provides a clear description of the conventional approach to calculating network energy using average intensity, but also describes an important new power model. See my discussion in Measuring website energy consumption via browser profiling.
- Clicking Clean Virginia, Greenpeace (2019) Greenpeace are often seen as extremists but they do a good job of raising awareness. In this report they attempt to estimate energy consumption of data centers in Virginia, US, particularly those used by cloud providers like Amazon, Google and Microsoft. They use FOI requests to obtain local permits for facility sizes and then estimate the total energy demand depending on the size of the data center. The energy supplier for Virginia is notorious for using fossil fuels and so Greenpeace does a good job of laying those alongside the “100% renewable energy” claims made by the big providers. See also: How can data centers use 100% renewable electricity?
- Digital technology and the planet, Royal Society (2020) A wide-ranging report from the Royal Society. Chapters 2 and 3 are of particular interest in relation to better disclosures and the impact of technology use.
Interesting tools related to sustainable computing:
- Carbon Aware SDK An open source CLI and JSON API to get carbon data into your applications. Written in C#.
- Cloud Carbon Footprint: This is an open-source software project led by Thoughtworks that connects to your cloud billing account and runs estimates on the kWh and associated CO2 from the resources you are using. I have made several code contributions to the core project, including implementing the cloud carbon coefficients methodology as a self-contained project that is regularly updated from server power consumption test data.
- CodeCarbon: Takes readings from AMD and Intel CPUs every 15s to provide energy consumption reporting mapped to carbon intensity.
- Coppers: A test harness for Rust that will measure the energy consumption of test runs using the RAPL API.
- Datavizta: Open database of carbon impact factors for computer equipment – laptops, monitors, servers, etc. Also available on GitHub.
- Electric Insights: Not sustainable computing specific but I use this on a regular basis to provide data about how the grid mix changes. It’s built and maintained by one of my supervisors at Imperial College London.
- EAM Environmental Modelling and Assessment Toolkit: Presented at the ACM SIGCAS/SIGCHI Conference on Computing and Sustainable Societies (COMPASS) and described in the related paper, but the repo is very sparse and has no instructions or documentation.
- greeNsort: A set of more efficient sorting algorithms that claims to be significantly more energy efficient. Lots of interesting reading on the website, but the code is not open source and is not available to download or use without buying a license.
- Scaphandre: An open source tool (written in Rust) that is designed to introspect power consumption from a bare metal host and/or Qemu/KVM VMs running on a host in a way that they can be graphed using common tools like Prometheus and Grafana. Only supports Intel CPUs.
- PowerJoular: Very similar to Scaphandre, but written in Ada and for Linux. Supports analyzing RAPL CPUs from Intel (Sandy Bridge onwards) and AMD (Ryzen onwards). Can also examine NVIDIA GPUs.
- Software Carbon Intensity (SCI): This is a specification for how to calculate a score that indicates the carbon intensity of software, produced by the Green Software Foundation. It’s not yet released as a stable spec and I have some concerns about how scores can be compared and why infrastructure measures have been excluded.
- Tools to measure software energy consumption: A good blog post that explains several methods for measuring how much energy is being consumed by your computer.
- Website power consumption measurement in Firefox (my blog post on this): Added in Firefox 104, you can now view power consumption stats on Windows (Windows 10 if there is an energy metering interface or Windows 11 via RAPL) and Mac (Apple Silicon only). See the implementation log and Windows and Mac changesets. This provides output in watts in the profiler UI, but the internals are measuring in picowatt-hour. On Mac it’s using the task_info API. Also worth mentioning the Firefox powermetrics guide which describes the measurement tools available in the Firefox source code.
See also the PyData 2022 slides from Green Coding which provide a good summary of energy measurement tools.
- Awesome Green Software: A list of dev/tooling, organizations and research/articles about how software developers can approach building software in a more sustainable way.
- Adrian Cockcroft’s talk at Monitorama 2022: Recently retired AWS VP, Sustainability, Adrian Cockcroft, gave a good 30min talk at the Monitorama 2022 conference: where he summarised the state of cloud carbon sustainability. Nothing new if you already know the topic, but a great intro to the current state of things.
- Hot Carbon 2022: A conference/workshop all about sustainable computing, with videos and papers for most of the talks.
- Guarantees of Origin, Renewable Energy Certificates and the Residual Mix in LCA: A good summary of the complications of measuring where renewable energy is generated and how to account for it. Relevant for understanding “100% clean energy” claims with PPAs and RECs.
- Open Sustainable Technology: A list of open tech projects, libraries, data, tools, and other useful things.
- Principles of Green Software Engineering: A set of principles for software engineers to keep in mind when writing software to try and reduce its environmental impact (which usually means carbon, but isn’t limited to that being the only measure).