Room 5 

11:40 - 12:40 


Talk (60 min)

Toward a carbon-aware Cloud

# "I have all my infra in the cloud, and I do not emit CO2."


The cloud holds a dirty secret: it generates greenhouse gas emissions on par with entire countries such as France or the United Kingdom (IEA, 2021). But there is hope! Infra & Cloud teams have the power to mitigate this issue. It all starts with a simple decision: choosing the right cloud region on platforms like GCP, Azure, or AWS. And guess what? This choice can reduce emissions by a factor of 10 or even 20! For instance, Ireland has a carbon intensity of 400 gCO2eq/kWh, while the French network can go as low as 20 g. I will demonstrate how we can measure and drastically reduce these emissions, making our cloud "carbon-aware."

In this presentation, I will:

- Address the issue of CO2 emissions in the cloud.
- Present the regulatory and financial landscape that compels us to address this problem.
- Introduce techniques to measure and estimate emissions using observability tools.
- Discuss practical methods to reduce the carbon footprint of your cloud infrastructure.
- Explain the concept of "carbon-awareness," where infrastructure dynamically self-adjusts based on emissions.
- Live demo showcasing a carbon-aware infrastructure.

# Who am I ?
Ex-first dev at Canva in Sydney and Elastic for several years, I have launched Carbonifer, a project aiming to make cloud infrastructures "carbon-aware."

Although this talk will not specifically focus on Carbonifer (no advertising), I will mention it among other tools.

# Why bother with all this?

Because the world is changing, and regulations and investors are urging action. National and European regulations are being enacted to ensure companies measure their carbon footprint. Companies will need to demonstrate tangible efforts, while ESG funds seek to invest in environmentally responsible enterprises. Moreover, adopting a sustainable approach serves as a powerful tool to attract talent and retain clients.

# Measuring

"You can't improve what you don't measure." - Peter Drucker

I will showcase the use of observability tools to estimate emissions in real-time for your cloud infrastructure. The carbon intensity of the electricity powering data centers is publicly available data, enabling us to make real-time estimations of cloud emissions.

# Reducing

Once we can measure, the next step is action. I will present practical methods to reduce the carbon footprint of your cloud infrastructure. We will go beyond simply pausing underutilised machines and explore options such as choosing low-emission regions. Additionally, leveraging auto-scaling groups and serverless services allows to keep emissions to a minimum.

# Carbon-Awareness

We can imagine an infrastructure that self-adapts based on its own emissions. When the electricity powering a region becomes too carbon-intensive, we can automatically reduce the number of machines in the cluster or temporarily relocate it to a region with lower carbon intensity.

To illustrate this concept, I will demonstrate how an infrastructure can seamlessly "move" a cluster of instances between regions based on the carbon intensity of the electrical grid, without any downtime or compromising service quality. Furthermore, we will explore intelligent scheduling of resource-intensive tasks, such as AI model training and video processing, taking advantage of regions with lower carbon intensity or favorable weather conditions.

All of this serves to demonstrate how infrastructure can dynamically adjust to emissions and become "carbon-aware"!

Olivier Bierlaire

Based in Nantes, France, I'm interested in sustainability of the Cloud and ICT in general.
Ex-founding team of Canva, I've worked at Elastic, and now focused on