Reading time: 10 minutes
This post was inspired by Asim Hussain's Principles.Green personal project and website, along with current events.
The energy crisis following Russia’s illegal invasion of Ukraine gave the world a wake-up call regarding its energy use (and waste.) This winter is shaping up to be challenging in terms of energy supply, with some operators and countries considering rolling blackouts or shutdown mandates.
As software engineers, architects, and managers, we can do our part to ensure the software we produce is more energy efficient: it’s better for the climate, the users, the bottom line, and the engineers (in no particular order!)
A good proxy for gauging carbon emissions is energy and resource use. Generally speaking, using more electricity and more hardware is correlated with dirtier software and higher emissions. By building greener software, the business also reduces its electricity, bandwidth, server, and hosting bills.
Optimizing software to reduce its carbon footprint yields better UX, as long as the UX designers are also involved early and make the right tradeoffs: less computationally intensive doesn’t mean bare bones!
Every optimization will help reduce emissions, but focusing on the wrong one isn’t the most impactful way to go. Without metrics and application profiling, the engineering team will be missing out on areas that need optimization the most and spend time on marginal improvements instead.
Once metrics exist, the organization can also use them to make better business decisions.
Designing software to report metrics encourages a more modular and flexible architecture which will be easier to maintain and extend for the lifetime of the software.
Having metrics also opens the road to monitoring, which is a significant contributor to better quality software.
Producing anything involves emissions, hardware is no exception. Once it has been produced, the carbon has been emitted whether the hardware is used for a couple of years only or longer; producing its replacement also emits carbon.
If hardware is replaced later, emissions are reduced: a 2019 Dell R640 server costs 1290 kg of CO₂ equivalent (kgCO₂e) to produce1. Replacing it after four years (Dell’s estimated product lifetime2) would have cost 323 kgCO₂e/year. But if replaced a year later, this number drops to 258 kgCO₂e/year.
Hardware that the organization doesn’t need anymore can be useful to other businesses or individuals. Almost every business I’ve worked with threw away vast quantities of 4–5 years old laptops, servers, monitors, furniture… Sending it to the recycler still means it will be destroyed (in most cases) rather than turned into a brand new item. A fraction only of its materials will be recovered; the rest becoming e-waste.
Instead, it might be worthwhile to pull the storage from it (which can be safely destroyed, or better yet safely erased) and sell the hardware. If the logistical burden on the organization is too much, donate the hardware to a local user group, or the friendly neighbourhood hackerspace. There are also companies specializing in collecting, refurbishing, and reselling decommissioned hardware as a service.
There might be tax implications to selling depreciated assets, but there are also ways around it; check with a CPA as this depends on your jurisdiction. The equipment could also be donated instead of sold, which can yield a tax benefit.
In a nutshell, power use doesn’t scale linearly with utilization: powering any server takes a nominal amount of electricity just so that the machine is on and ready to work. So, to do nothing (0% utilization), an example server would use 250W. At 50% utilization, that same server might use 280W; and to run a 100% load it might use 295W.
Thus, reducing the number of individual machines and consolidating their load onto larger machines to make them busier will use less power overall. To further optimize power usage, CPUs could be made to run slower off-peak load hours. This is where the metrics collection facilities mentioned earlier are also useful.
Similar to how appliances have an “eco” program or mode that makes trade-offs between resource use and performance or time, software could also have an “eco” mode. Every major OS has a way to adjust computer or smartphone performance to reduce energy use (and prolong battery life).
As far as I know, this is a largely untapped area for SaaS or local software: I don’t know of any application or web app that lets me adjust performance to reduce resource use directly. Most SaaS will offer different tiers but they’re about market segmentation and revenue optimization, not directly to reduce carbon emissions. One could also imagine a monthly discount based on how much time was the web app running in “eco” mode as it would translate to lower variable costs for the provider.
The toggling could also be automated so that the application runs in “eco” mode during periods of peak dirty electricity generation and in performance mode during clean energy surpluses.
Along with having an “eco” mode, software could be architected to run its heavier computations at times and in locations where there is a surplus of clean energy, as it’s not always necessary to run computations at the nearest edge and instantly. The goal is to take advantage of a clean energy surplus as it can’t easily be stored for later use, contrary to computation results which can be saved and recalled later.
Moving data around incurs a carbon cost, both in electricity to send the packets but also in the extra hardware required for the extra traffic. By sending less data over shorter distances, not only is less carbon required for transmission, but it also reduces the hardware requirements to accommodate the traffic. And it results in less data to process for the client, which might also reduce power usage. This can be achieved in practice by using CDNs.
Applications might also optimize their connections and keep them open longer rather than closing them each time. Protocols like HTTP2 enable this3: it reduces the number of handshakes required, resulting in lower traffic and computation requirements.
Reducing network use might sound contradictory to computation shifting, and it is. Each organization should consider the tradeoffs of each strategy and decide on the best mix for their use case.
A performant application is an optimized application. And an optimized application uses fewer resources to achieve similar or better results than its unoptimized equivalent. Having application performance as a core requirement will yield lower carbon emissions (and higher quality software overall).
Depending on the current organizational culture, this will require more or less effort but it will organically enable most of the other techniques mentioned in this post.
While this post mostly focused on software architecture and hardware so far, there are things organizations can do beyond software that contribute to reducing overall emissions; setting an overall tone and philosophy for the way an organization is doing business.
Incentives through benefits are a financially efficient way to increase compensation and attract candidates, so why not encourage individuals to make greener choices? Organizations can subsidize the cost of better transportation choices (buying and using a bicycle or using public transit instead of driving), implement partnerships with power companies for a discounted cleaner energy rate at home, incite repair and reuse of company equipment, reduce meat and fish consumption, consume locally produced foods…
Most utility companies offer cleaner energy plans, so why not switch offices and data centers to these instead? It’s usually rather straightforward and encourages utilities to build even more renewable energy generation capacity.
A step further than switching to clean energy is generating it. Solar panels are subsidized in many municipalities and pay for themselves rapidly. Efficiency has drastically improved over the last few years, and the cost has gone down drastically.
Rooftop food gardens, bee hives, rainwater collection systems, or simply covering in grass will reduce the business’ carbon output, mitigate urban heat, and help biodiversity. Organizations can use services that specialize in looking after these installations for a hands-off implementation.
After years of pandemic, remote work has become mainstream. It is not only a way to reduce emissions from healthcare (which causes between 1% and 5% of global emissions4) through lower rates of infectious diseases and lower stress levels from not sitting in traffic, it also reduces energy expenditure for office space and personal transportation.
Not every employee enjoys remote work, so a hybrid model where everyone is free to choose for themselves how much or how little they want to be remote might make the most sense.
Flying is notoriously polluting and isn’t always necessary. Meetings don’t always need to happen in person across long distances and can sometimes be done via video call.
When people do need to meet, consider trains as an option. Trains don’t always make sense (long distances, trans-oceanic travel), but it’s an often overlooked option even when it would make the most sense. Trains can also be cheaper and a less stressful experience than flying: travelers don’t need to get there 2 hours in advance, no need to wait for luggage, and it goes from city center to city center directly.
Along with remote work, this is becoming less and less of a radical idea: several companies in the UK have recently switched to 4 days a week at 100% pay for 100% productivity5. Working one day a week less improves quality of life, which in turn reduces stress and healthcare needs but also results in better output from employees6.
Making greener software requires more optimized and better-architected code which is more likely to happen when workers are healthier and happier.
The climate crisis has been looming for decades without much action7, the Russian invasion of Ukraine shook things up further by drastically raising energy prices. Energy shortages could be part of our near future: building software to use less energy and do it more efficiently is part of the solution.
Besides making economical and environmental sense, it is also positive for an organization’s public image. Generation Z is particularly sensitive to environmental issues8, which influences their consumption choices.
Furthermore, we are likely entering an economical recession, and cost reduction makes even more sense in that context.
I hope some of the strategies in this post will inspire individual readers to advocate and push for change in the way software is built. Greener software is an emerging trend, and now is a good time to get ahead of the curve.
https://i.dell.com/sites/csdocuments/CorpComm_Docs/en/carbon-footprint-poweredge-r640.pdf or https://archive.ph/YLvgJ, page 1, 7730 kgCO₂e × (16.6% from manufacturing + 0.1% from transportation) = 1290.91 kgCO₂e ↩
https://i.dell.com/sites/csdocuments/CorpComm_Docs/en/carbon-footprint-poweredge-r640.pdf or https://archive.ph/YLvgJ, table on page 2 ↩
https://web.dev/performance-http2/#one-connection-per-origin or https://archive.ph/Fw11Q#one-connection-per-origin ↩
https://www.thelancet.com/journals/lanplh/article/PIIS2542-5196(20)30121-2/fulltext or https://archive.ph/Bvr1N ↩
https://edition.cnn.com/2022/06/06/business/four-day-week-trial-uk/index.html or https://archive.ph/2Vnob ↩
https://www.bbc.com/worklife/article/20210819-the-case-for-a-shorter-workweek or https://archive.ph/D1YVa, “Evidence suggests that one of the biggest advantages of working fewer weekly hours is that it makes people better workers.” ↩
https://www.theguardian.com/science/2021/jul/05/sixty-years-of-climate-change-warnings-the-signs-that-were-missed-and-ignored or https://archive.ph/G8j6r ↩
https://www.pewresearch.org/science/2021/05/26/gen-z-millennials-stand-out-for-climate-change-activism-social-media-engagement-with-issue/ or https://archive.ph/GMIH0 ↩