Transcript
Ludi Akue: Two-thirds of the world’s disasters are natural disasters: earthquakes, landslides, storms, floods, extreme heat. In 2022, here in the UK, because of extreme heat waves, two major NHS hospitals suffered IT failures. This impacted patient care. Natural disasters are getting worse, more frequent, more intense, and they are affecting everything, including digital systems. According to Paris Agreement, this will get worse if we don’t limit the global warming to 1.5 degrees Celsius. To do that, we need to reduce by 45% by 2030. We need to reach net zero global emissions by 2050. We have a problem. Do you know why we have a problem? We have a problem because the tech sector’s emissions is still increasing. The tech sector is 6% of the global emissions.
When I talk about emissions, I’m talking about greenhouse gas emissions. This surpasses the whole airline industry. Because of AI, emissions are estimated to triple by 2030. We have to do something, if not, we won’t meet Paris goals. Natural disasters will become uncontrollable. I wish I knew all that when I started working with Green IT, that is four years ago. I was the Chief Technical Officer of Lunii, a hardware scale-up in France, a French market leader for kids’ toys.
In France, I led Green IT transformation. I wish I learned that four years ago. Currently in my mandate as the Chief Technical Officer at Bpifrance Digital, I’m still working on those issues because it’s our central role as tech to do something. If not, we won’t reach the 1.5 degree Celsius limit. I want to share with you the seven lessons I learned the hard way. To get started, Green IT is strong and avoid common traps. Let’s go. One thing I want to make sure, I think, is to be all aware that we are now building tech in a new operating context, an operating context of a destabilized world. This is our new design constraint. Emissions are design constraints. Planet boundaries are design constraints. Regulations are also design constraints.
Lesson 1: Assessment Challenge
Let’s start with the first lesson, where to begin. Assessment is very challenging. When we started at Lunii, we had no clue where to begin. We were a hardware company. We make connected toys for kids. We published audiobooks as well. We are operating in Europe and starting to operate in North America. As we didn’t have a clue, we brought in a sustainability expert. You have a lot of concepts to learn. It’s not overwhelming, it’s ok. The first concept is LCA, Life Cycle Assessment. It is the standard method to get started with not only Green IT, with sustainability. You just review your whole supply chain from raw materials, manufacturing, usage, and beyond to evaluate your emissions. This is the start. You have to learn also our scopes. Is anyone familiar with scopes here? Scopes are frameworks to categorize emissions. You have scope 1, which is emissions from your activities.
For instance, you have a company delivery van, it’s scope 1. You control scope 1. Scope 2 is emission from energy you use for your activities. For instance, your office electricity. Scope 3 is the rest, the whole supply chain you are operating from your logistic partners to your cloud provider. You also need to have a goal. Because without a goal, you cannot improve anything. Actually, we chose to measure our carbon footprint. A trap here is to feel overwhelmed with all the discrepancies you can have in measuring the carbon footprint, because all the numbers you will get will be wrong. That’s ok. That’s ok because measuring carbon footprint is a relative activity. You can measure something as a declarative and some cannot be measured. That’s ok. Feel free with that. Don’t look for clear measures, just have a sense of direction to get going.
A trap to avoid is it’s not just our carbon, because tech induces stress on materials, resources like water usage, stress on the power grid. You have to start somewhere. Start with carbon. Carbon is cool. Carbon is simple. It gets you going to create momentum. I started with carbon with my team. When we did the LCA for the whole company, what we learned is that tech was the third contributor to the global emissions of our company. We choose to look at scope 1 and scope 2. Scope that we can control. In the tech contribution, you have everything IT related: devices, screen, laptops, servers, and you also have infrastructure. The first thing to do is to get in really with infrastructure.
Lesson 2: Infrastructure Reality
How many of you are on the cloud? On-prem? A mix of both? I was on a Platform as a Service, the major one. I used to think that because I’m on a PaaS, everything is ok. What I found out is the PaaS is the least flexible and the least transparent cloud infrastructure because the abstraction layer, it adds to the cloud. Everything is opaque. The scaling is opaque. You cannot fine-tune your own database. You cannot look at what is working under the hood. Because everything was difficult to measure, I decided to move to the cloud, partly to gain control. The lesson here is simple, you have to gain control and push for more transparency. We moved to the cloud and we learned a lot of things.
One thing you need to understand is what’s called electricity mix. Electricity mix is, what is the composition of electricity in your country, if you are on-prem, or in your cloud regions, if you are using cloud? For instance, electricity in France, because of nuclear, is green. It’s less carbon intensive like electricity in Virginia. When we were on the PaaS, we were located in Virginia. Our region was Virginia. The first thing is to understand the electricity mix and the carbon intensity. One thing also to understand when you understand the carbon intensity, is that you can shape your workloads.
If you are on the cloud, of course, you can move your workload to the greener regions. Let’s look at operational batch jobs, for instance, during low carbon intensive hours. It’s the new skill to build to understand how your demands are stressing the carbon intensity, your power grid. One thing that helps, and I think the major cloud providers did a great job is that they all have sustainability guidance. Please follow them because it’s well done. If you are on-prem, you can look at those sustainability guidance and apply it also to on-prem. If you have the chance to improve the hardware, always look for efficient hardware. ARM, for instance, are very efficient and less costly for the planet.
Lesson 3: Software Architecture Paradox
This makes me come to lesson three. Lesson three is about architecture. Picture this. In the company, we had a microservice setup. If I can share some requirements of how we are doing, it’s very simple. We build hardware. We track logistics. We publish audiobooks in 13 different countries, so multilingual audiobooks. We build e-shops, e-commerce shops to sell audiobooks and to sell hardware. Those are the big requirements of our architecture. At that time, you build everything in-house. Looking through the Green IT lines and using DDD, I make three moves. The first one is simplify. We are overbuilding everything in-house. We are scaling and going to North America. We have a lot of tech debt piling up. We need to make a choice, have a focus. What is our core domain? Then use the core value of the business.
At that time, it was audio publishing. Then, what is our supporting domain? The supporting domain like e-commerce. It’s ok to delegate this part to e-commerce like Shopify and stuff. I chase friction because when you’re looking at technical debt, technical debt signals opportunities, Green IT opportunities. First, simplify, core domain. I made a choice to help the team succeed. It’s to go from microservices, which because of the technical debt is becoming a distributed monolith, back to modular monoliths to make things simple. That’s the first thing. We migrate all of our e-shops on Shopify. The third thing is, I refocus the whole team, sociotechnical, I refocus the whole team onto the core domain, bringing simplicity. It’s correlated with Green IT.
Lesson 4: Frontend Migration
Let’s go to lesson four. Lesson four is really funny because frontend is the most overlooked when it comes to Green IT. Fun fact, webpages are energy and emissions. This work from HTTP Archive says a lot. In the last 10 years, the weight of our webpages grew by four times. That’s a lot. What we don’t see is that the heavier webpages strain users’ devices. Make users charge more and recharge more. Force the user to change each device to compete with the complexity of the webpage. This is something to keep in mind also. Of course, when we started with frontend, we were like, we need to optimize page speed. Because page speed correlates with conversion rates. We were into page speed. We were also into accessibility. We were on all that, and we started to look at Lighthouse to improve our core vitals.
Then I discovered a new world of page size, weight. Also, when you throttle the network, you can see how your website is behaving in low network regions. You see that webpage weight is not just tech, it’s not just bad UX. It’s an equity issue. We have a ton of tools to optimize for performance. When I talk about the webpages, I’m talking about HTML, CSS, JS bundles, video, image. We must compress images. We must lazy load, defer the scripts. We must trim them. There’s a lot of resources to do that for free on the web. Don’t forget webpage optimization. Also, if you can afford it, go for static rendering and edge caching, to be close to your user and to limit the data transfer. Data transfer is very expensive when it comes to emissions. Use efficient protocol. You have HTTP/2. Now we are going to have HTTP/3. If you can afford it, go for it. Keep measuring your real frontend impact.
Lesson 5: FinOps Correlation Trap
This one is also funny, because when I talk about it to our operation teams and DevOps teams, the thing that’s come up even with CFO is, FinOps correlates with Green IT. Yes, until it’s done, it doesn’t. Why? Don’t assume cloud is always greener. I’ll tell you why. Let’s take a simple example of FinOps. Let’s go for cheaper regions. Cheaper regions are often carbon-intensive regions, that’s why they are cheap. When you move your workloads to cheaper regions, you are worsening your Green IT endeavors. I want to speak about, for instance, Instance Spot. Who uses Instance Spot? Who knows about Instance Spot? Instance Spot is a service that you can ask.
If you are doing Green IT or FinOps, you prefer to use what you need, Instance Spot is here for that. You can just ask for an instance when you need it. This appears very green, but if you look under the hood, Instance Spot used to be old virtual machines. You think that you are doing FinOps. You think that you are doing Green IT. Under the hood, you are making things worse. The thing is, you need to rethink your decision and you need to be curious to see what is working under the hood, even if you are using a cloud. For your FinOps to work, you need to include in your FinOps, impact metrics, and to measure them. You need to measure the electricity. You need to use your cloud calculators. It’s a whole new subject. That’s what we did at Lunii. You need to build walkways in your procurement.
Lesson 6: Enabling Change
The lesson says, as QCon folks, we all know that it’s sociotechnical. It’s not because you do LCA, Life Cycle Assessment. It’s not because you choose a goal, like to measure your carbon footprint, that it will work out of the box. You need to enable change. I prefer the term enabling change than managing change. How can you do that? What we did and what some of the companies do is to build external accountability. It’s like, if you decide to lose weight or to go to the gym, you tell your friends, “I decided to go to the gym two times a week”. You build external accountability to help you when things become hard, not to give up. This works. Even as Bpifrance, we build external accountability as well.
The external accountability of Bpifrance is we are the climate bank, that is one of our core missions, is to accelerate the green transition of the French economy. You start with building accountability. When we are talking about Green IT, it’s very new. Someone was telling me it’s niche. You have to train the people. You have to educate them. They need to learn planet boundaries, system thinking, new reflex to unlearn some reflex performance, optimization reflex, to learn about sufficiency, not just efficiency. When you create this space of learning and iteration, you want to build in feedback loops. It’s not a 3-year goal, like I want to build in Green IT and full stop.
You have to put milestones, metrics, celebration, and a space for the team to reflect on the challenge. Sustainability cannot be a side quest. I made that mistake. What I learned is you need to build sustainability in your delivery backlog. It’s just another user story, like your whole user story in the backlog. You need to have the company’s OKR translate sustainability. This will be easier to do than to have sustainability in the tech team. Don’t centralize mandates. Build a distributed leadership team. Appoint an enablement team of champions in all the teams: tech teams, product teams, design teams, marketing teams. Let the champions build the change together. That’s what I did when I was at Lunii. We have the champions. It’s very difficult to build the change, to enable the change, so distribute the leadership.
Lesson 7: Emerging AI Patterns
The final lesson is not a story of the past, it’s my current work. As a member of the Green IT working group at Bpifrance, we are looking at emerging AI patterns when it comes to sustainability. It’s not just that I think, nobody is talking about sustainability at the governance level right now. Let’s take an example. Even if you have the AI Act in Europe, the AI Act barely touches the question of sustainability. At the same time, AI is accelerating everywhere. We are talking about agentic AI, so embedded AI everywhere. We are watching a lot of models coming to life and dying right away, with heavy training costs. For the first time, it’s not today, it’s like many months ago, inference is the new energy thing. That is, inference surpasses training costs in terms of emission. Not to mention the hardware trained with. Do you know that GPU chips, their lifespan is about two, three years?
If you are lucky, five years. Nobody is talking about that. Nobody is telling the users, is telling us, what it costs today to have AI everywhere. There are a lot of bans in different countries against the building of new data centers, because the data centers are stressing the local power grids also. We need to do something about that, because everything is free, but we know that it’s not free. No pricing means no restraint, and our product and tech teams are building AI tools without ever thinking about, do I really need to use AI here? How can I build in environmental telemetry in the AI usage? How can I inform the users to make the right choices? Nobody is doing that. I’m trying to do that right now.
One thing we are doing at Bpifrance is to promote AI literacy. I work on PromptSage GPT, which is a custom GPT that you can try to learn how to prompt better with the tool. The tool will help you prompt better and build sustainable prompts and traditions as well. What I’m trying to say here, is there are many tools to achieve telemetry, to achieve the control of our usage of tokens, because token usage is emission. You can refer to tools like EcoLogits, maybe some of you know about the tools. EcoLogits, you have LiteLLM and you have Langfuse to monitor your usage of AI.
Key Takeaways
When I began this journey, I wish I knew all that. I wish I knew that we are operating in a new operating system of a destabilized planet. I wish I was aware that tech is playing a central role. As software engineers, tech leaders, we are playing a central role in bringing the global warming to 1.5 degrees Celsius. If you have one thing to keep from this talk, it is, you have to learn a lot, yes, but you have to unlearn a lot also, and build a new set of questions. Why, is the most important one. Why am I building this? What am I building? Where, how, and where am I running it? For whom am I building this tool? To build an intuition for sufficiency, that’s very important. I will leave you with three pieces of advice.
The first one is, resources are abundant everywhere. It’s not a matter of tooling, it’s a matter of space and reflection. First, be curious. Be curious to go out of your comfort zone, because Green IT requires to go out of one’s comfort zone. The second one is honesty to assess our impact, and to own our negative personalities that we offload for years. Three, we need to collectively have this courage to push our cloud providers, SaaS, AI, for more transparency. Happy Green IT journey.
Questions and Answers
Participant 1: Could you elaborate a bit on why data transfer is such a large component of emissions?
Ludi Akue: It’s not the largest, it’s the third largest. This is typical as a hardware company, the production was the first contributor, because we are having chips from countries and producing the toys for the kids. The second one was the distribution, that’s the logistics. That’s normal. It is very common to have this kind of setup. The tech, they come in second or third row. When we measure our evaluation from scope 1, electricity, office electricity, and scope 2, what we’ve learned is that it’s more the IT side that makes the whole contribution go up. IT side is per employee, so you have the laptop, the mobile phone, you have the servers, you have the networks. This is the grid part of our carbon emission.
Participant 2: One of the slides mentioned imperfect data. I wanted to hear about your experience of measurement, also scaling that measurement and handling imperfect data and making people believe it.
Ludi Akue: I think that you have to be comfortable having false data, have relative data, because you need to get going. We started like three years ago, with a tool called Carbo. Carbo is a French tool. Carbo, you cannot measure everything. Because, for instance, you have your application running on the cloud. To calculate the emissions of your application, you have to rely on the cloud calculators. For instance, Google Cloud has what we call the local based calculator, that is your actual emission according to your region, the region you are using. AWS for instance, has market based, so it’s unclear. You cannot have the finest number. You have to be ok with that. The numbers vary between tools, also. I told you about Carbo, it’s an example.
Let’s take a bigger company, because at Bpifrance we are more than 5,000. I have 20 applications, and some parts you have maybe 400 applications. To measure that is very difficult. You have cloud, you have on-prem. How do you do that? You need to ask the data center provider to give you the calculations. How can you measure like your webpages? You have to go with relative numbers, but it’s ok. Because you are looking for a sense of direction. What is more important is to have a method to do that. We are still thinking about a good method at Bpifrance right now. It’s ongoing work, and that’s ok. It’s Green IT, it’s not simple. We need to learn. In three years, I can see shifts in the tools. There are many tools. I will complement my talk with a document about the tools that were available. Things are getting better, but don’t look for the numbers, it’s not the real numbers. Look for the sense of your numbers to get going.
Participant 3: How much convincing does it take from the business side? Because I’m guessing you can’t just be like, this is for saving the world. They might want to be more convinced on the money side of things.
Ludi Akue: You get it right, it’s not just about saving the world, I think it’s for resiliency. Don’t forget the initial message. I think it’s very important to acknowledge that we are working in a destabilized world, and that natural disasters are causing supply chain disruptions, are threatening lives. When you come to your companies, it’s to let them know, it’s not easy. You have to build a business case around that, and you have to build a business case around looking for external accountabilities, even if it’s for, not greenwashing, but it’s for communication, it helps. Looking to find how your heavier webpages are slowing the business, are slowing your scale, and looking for business disruption signals. Tech debt signals efficiency, you can do Green IT through tech debt as well. If you have friction somewhere, just look at the friction to build Green IT cases. Have champions with you, like talk to your early adopters.
Participant 4: How do you help dev teams to spend time on Green IT subjects? Do you have dedicated Jira tickets in the sprint backlog, or weekly meetings to explore metrics, or stuff like that?
Ludi Akue: It’s different. Actually, I tried, ok, do Green IT. The champions have the team come up with user stories, so Green IT user stories. I tried to let them do the stories when they have time, and this is a huge mistake. I decided that they will have 20% of their time for the Green IT. It’s an option, but if you don’t have this option, just make sure that every team embarks on at least one Green IT user story in their sprints, if they have sprints. I think the trap here is to make it a side quest. You want to build a new normal behavior. A new normal behavior is, please be alert, be aware of Green IT optimization, and build that into your backlog. Don’t wait for team gathering. Let’s celebrate reaching a Green IT goal, also. What I hope is, in a few years, we will forget about the Green IT side of the equation. It’s just normal work, but we are not there yet.
Participant 5: Building on your question which is around enticing leadership on this journey. Really, what I want to know is, what advice do you have around bringing developers on this journey and enticing them? Also, is there a way to wrap in regulation on top of this?
Ludi Akue: The chance I have is to have young developers. Young developers want to change the world. I don’t say that other developers don’t, but young developers are ready for that, are ready for the change. Something that’s very interesting at Lunii, I talked about Green IT, but Green IT is a part of the whole sustainability movement. Green IT is tech, but when you think of circular economy, it’s part of the sustainability movement. The younger generation is ready for that. They need methods, approaches, and create time for them to do that. For people not ready for that, culturally not ready, it’s a question of education. We have a sustainability leader at Lunii, and her goal was to train everyone on sustainability, on Paris Agreement, on how to measure, and to make everybody easier with the fact that we don’t know. We estimate, but we don’t know, we need to get improving, and we don’t know the real answer, and to get cool with that. I think that education is a part of that, and maybe meeting with other teams doing sustainability work, BBI, things like that.
Regulation is interesting, because I can build up on the leadership thing. Regulations are here. I think the most constraining regulation — I’m in fintech right now — you have to disclose your climate risks. It’s a lot of reporting, but you can use it as a lever to start a Green IT journey. The climate risk, you have ESG, you have those reporting this, and many of them are the whole one. Use the regulations, so this kind of regulation that is constraining us to report our carbon emission, is constraining us to report our sustainability, the global sustainability, equality, use that to foster a sense of movement, a sense of transformation. Where regulation is not helping right now, for me, is in terms of AI and sustainability, and we need to build that kind of governance.
See more presentations with transcripts
