Minimising the Carbon Footprint of Workloads

What if your SLOs were based on carbon emissions instead of latency? Learn to build carbon-aware software and reduce your workload's environmental impact.

#1about 2 minutes

The growing carbon footprint of the IT industry

The IT sector contributes 4-5% of global carbon emissions, a figure larger than the aviation industry and projected to triple.

#2about 3 minutes

How AI workloads accelerate energy consumption

Both training large models like Llama 3 and running inference for services like OpenAI consume massive amounts of energy, driving up emissions for major tech companies.

#3about 3 minutes

Emerging regulations for data center efficiency

Governments are beginning to regulate data center energy use and grid strain, but a general lack of awareness and transparent data from providers hinders progress.

#4about 5 minutes

Key concepts for sustainable computing

Understanding server energy proportionality, Power Usage Effectiveness (PUE), and the embedded carbon from hardware manufacturing are foundational to reducing IT's environmental impact.

#5about 3 minutes

Practical strategies to reduce workload emissions

Simple but effective measures like eliminating zombie servers, right-sizing instances, using auto-scaling, and adopting ARM CPUs can significantly lower carbon emissions and costs.

#6about 1 minute

Tools for measuring energy and carbon emissions

Open-source tools like Kepler for Kubernetes and Scaphandre for Linux can measure energy consumption, which can then be converted to carbon emissions data.

#7about 2 minutes

Tracking emissions with Software Carbon Intensity (SCI)

The Software Carbon Intensity (SCI) ISO standard provides a formula to create a score for your application, which can be used as an SLO to prevent regressions in your CI/CD pipeline.

#8about 4 minutes

Using carbon awareness to shift workloads

By understanding real-time grid carbon intensity, you can time-shift batch jobs to sunnier hours or region-shift development workloads to greener data centers.

#9about 3 minutes

Case study on optimizing a GKE cluster

An experiment deploying a microservices application on GKE demonstrates that tuning default resource requests and enabling auto-scaling leads to optimal server utilization and lower energy use.

#10about 2 minutes

Green coding and on-premise optimization strategies

For on-premise environments, consolidate workloads to power down unused nodes, and at the software level, focus on profiling to find and fix inefficiencies rather than just changing programming languages.