David Mytton

Sustainable computing – where to focus?

Published in Data Center Energy, Data Center Water, Environment, IT Energy.

Sustainable computing – where to focus?

Climate change is the biggest and most important long-term challenge we face. Everything about tackling this problem is focused on carbon dioxide (equivalent) emissions and how we reach net zero, or ideally net negative.

The underlying story in the Dispatches documentary episode I was recently interviewed for was that the tech industry uses a lot of energy, which is generated from sources with a high carbon intensity. Such analysis typically focuses on the use-stage energy, usually electricity, because renewables still make up a minority of global primary energy demand. As of 2019, wind power accounted for 5% of global power supply, solar 2.5%, and the rest is either natural gas, oil, or coal, with a small amount of biomass and nuclear.

Framing the problem in this way allows the media to position big tech as part of the problem because they run the infrastructure behind an increasing proportion of the services we all use. Energy is a major component of this environmental impact, and is one of the larger causes of the carbon emissions contributing to climate change. However, it is not the only environmental impact of computing and there is a risk of tunnel vision if we only focus on carbon.

Moving from coal power generation to wind power not only reduces the carbon intensity of the electricity generated, but it also reduces air pollution, decreases water intensity, and removes the need for mining. These are all important environmental improvements, but they have become secondary with the focus on carbon.

Energy and sustainable computing

Data centres get significant focus in discussions about sustainable computing because it’s easy to pick on an industry that is growing rapidly, consuming a lot of energy, and where the biggest player (Amazon) lacks transparency. This seems intuitive because we all get an electricity bill and can link charging our laptop or phone to what we pay each month. However, the tech industry has made significant progress in decarbonising the infrastructure it owns – mostly data centres. When it comes to the big cloud providers, this narrative is becoming outdated and at least Google and Microsoft are fighting over who has the greenest cloud.

Yet despite this progress in the data centre world, there remain some major areas that need work. Whilst using public cloud in super efficient hyperscale data centres might be standard practice for modern organisations and startups, there are questions around how rapidly workloads are migrating from older data centres.

Small data centres are another area of uncertainty – with just a few servers, they still make up 40% of all data centres in the US. These facilities are very inefficient and consume 13TWh of energy annually. This is between 2.5-6% of total data centre energy consumption (depending on which estimate you use) but the carbon impact is still meaningful.

Then there’s everything outside of what the public cloud providers own, but that is still within their supply chain. Google and Microsoft both provide detailed environmental reporting for their own data centres but what about everything else? For example, Google publishes figures about 17 of its data centres but note how many locations are missing. Where are the detailed about its London cloud region? Or Tokyo? Either these are outside the scope of “large scale” and/or “campuses with at least twelve months of data”, or they are being excluded because Google doesn’t own them. Perhaps Google is leasing space in a colo provider instead, thereby outsourcing its emissions to a third party.

What about the network? When you stream a video from YouTube, you are responsible for the energy consumption of your laptop and Google is responsible for the energy consumption of its data centre, but what about the infrastructure inbetween? You have a home internet provider but that connects through multiple networks before it reaches Google. Network energy efficiency improves by 50% every 2 years, but even so data transmission networks consumed 250 TWh in 2019, more than data centres. Global internet traffic increased by 40% between Feb and Apr 2020 due to the Coronavirus lockdowns. This is the subject of an upcoming paper I expect to publish next year.

Graph to show estimates for electricity intensity for the transmission network system from Aslan, et al (2018).
Graph to show estimates for electricity intensity for the transmission network system from Aslan, et al (2018).

Materials and sustainable computing

The big players like Microsoft, Google, and Facebook, have been successful in their transition to powering their data centres with renewable electricity. This means that even as data centre energy consumption increases, the carbon intensity decreases.

Carbon footprint depends on more than just energy consumption. Although the energy consumption of Facebook’s Prineville data centre increased between 2013 and 2019, its operational carbon output decreased because of renewable-energy purchases. From Gupta, et al (2020).
Although the energy consumption of Facebook’s Prineville data centre increased between 2013 and 2019, its operational carbon output decreased because of renewable-energy purchases. From Gupta, et al (2020).

This is great progress, but computing is much more than just energy. There is the water required to generate that energy, and cool data centres, but also all the materials needed to manufacture, transport, and recycle.

The hardware life cycle includes production, transport, use, and end-of-life processing. Opex-related (operational) carbon emissions are based on use; capex-related emissions results are from aggregating production/manufacturing, transport, and end-of-life processing. From Gupta, et al (2020).
The hardware life cycle includes production, transport, use, and end-of-life processing. From Gupta, et al (2020).

Behavioural changes are often what people focus on when thinking about how they can combat climate change. Eating less meat is a good idea. Cycling is better than driving. But these types of interventions get too much attention. Most people are not going to shift their behaviour. They are going to continue listening to music using videos on YouTube despite it using significantly more energy than audio-only streaming. Or accept the Netflix default of streaming in 4k quality even if they don’t have a 4k TV or laptop. When it comes to sustainable computing, the best way to reduce your impact on the environment is to not buy a new laptop or phone. If you buy a new iPhone 12, only 14% of the carbon footprint is in using it. Whether you charge it using renewable electricity hardly matters when 83% of the carbon goes into the manufacturing.

The material aspects of computing need more attention – mining rare earth metals, chemical processes to produce circuit boards, manufacturing of semiconductors, construction of factories and other facilities like data centres, mass production of electrical components, wires, LCD panels, printing packaging, transport around the world, recycling methods and reclaiming used materials. These are key components of detailed Lifecycle Assessments, yet we rarely have sufficient insight into the supply-chains of major electronics and computing manufacturers.

This is why more pressure needs to be placed on the organisations we transact with. Apple is a leader with their commitment to using 100% renewable electricity by 2030 and the efforts it has made in using recycled materials. But they get a big FAIL when it comes to repairability. It’s better to be able to buy a small replacement part than to have to ship your laptop off to a workshop, which due to how Apple designs its systems often requires replacing many large components.

We can signup with an electricity company at home that contracts to provide us with renewable electricity, and we know that Google owned data centres are powered by renewable electricity (on an annual basis), but everything else involved with streaming of a YouTube video remains a black box. We are only at the beginning of understanding the true environmental impact of computing.

Sustainable computing – where to focus?

The transition to renewables is happening. Renewables will account for almost 90% of the increase in global total power capacity in 2020, and this will accelerate in 2021. There is still a long way to go but the economics of wind (and solar) have been proven.

Data centre energy is following this trend. Microsoft and Google are making the most progress but Amazon has also pledged to follow them. Colo and traditional data centres need to work a lot harder, and there has been little focus on the networks connecting the user to the services they use.

Progress in renewable energy has pushed IT carbon footprints down, particularly in the use-stage, revealing how much more work is needed in manufacturing. More attention needs to be paid to the rest of the supply chain. Data centre water consumption is a grey area, and lifecycle analysis needs to expand from just energy to examine the full environmental footprint of computing. There is a lot more research to be done!