Green Software - Introduction to Carbonaware SDK and Carbon CI p... Szymon Duchniewicz & Fergus Kidd

### Overview Presented by Fergus Kidd and Szymon Duchniewicz from the Green Software Foundation (GSF)—a non-profit under the Linux Foundation umbrella—this talk introduces key initiatives aimed at reducing the carbon footprint of software systems. The speakers focus on two major foundation projects: the Carbon Aware SDK and the Carbon CI/CD Pipeline. The foundation works to build a trusted ecosystem of standards, tools, and best practices, encouraging culture change across the tech industry toward environmentally conscious development.

### Key Topics - **Sustainability Scope and SCI Specification**: Distinguishing between sustainability *with* technology and sustainability *in* technology, the talk emphasizes measuring software carbon emissions. The foundation developed the Software Carbon Intensity (SCI) specification, a standardized methodology to calculate the rate of carbon emissions for any software system based on defined software boundaries, stable scale metrics (e.g., per API call or per user), quantification methods, and reporting. - **Carbon Awareness Principles**: Carbon awareness centers on scheduling or routing compute to do more when energy is clean (high renewable mix like solar or wind) and less when it is dirty. The primary strategies are *time shifting* (delaying asynchronous workloads to cleaner hours) and *location shifting* (routing workloads to regions with cleaner energy grids). - **Addressing Industry Pain Points**: The talk outlines obstacles such as lack of unified tooling, reliance on varied carbon data providers, regulatory pressure (such as EU compliance), auditability concerns, and risk of greenwashing.

### Technical Details - **Carbon Aware SDK Architecture**: Built in .NET with Swagger and OpenAPI specifications (enabling client libraries across up to 50 languages), the SDK sits between cloud/local applications and external grid emission providers like WattTime and Electricity Maps. It abstracts API variations, manages API keys securely, and offers endpoints such as `emissions/bylocations/best` and `emissions/bylocation/bytime`. - **Jupyter Notebook Demonstration**: Szymon demonstrates an intensive daily data processing job running in Azure's West Central US region at 8:55 PM (928.5 g CO2/kWh). Shifting execution time to 11:55 PM yielded a 47% carbon reduction for that specific day (and an average 38.76% reduction over a 55-day historical evaluation). Shifting location at 11:55 PM to Azure region `West US 3` (mapped from WattTime's `azps`) delivered an additional 30.3% single-day reduction (18.18% over 55 days). Shifting globally to France Central (utilizing high nuclear energy) achieved an overall 51.76% reduction over the original baseline, demonstrating overall application carbon savings of up to 76%. - **Carbon CI Pipeline Tooling**: Fergus explains that post-deployment retrofits are reactive. The Carbon CI pipeline integrates an explicit "Assess" phase into standard Code/Build/Test/Release workflows. By forecasting SCI prior to deployment, engineering teams can proactively evaluate and adjust deployment regions, timing, and resource scaling before infrastructure generates carbon.

### Takeaways Carbon-aware computing offers substantial immediate reductions in emissions through practical workload scheduling and region selection. Tooling like the Carbon Aware SDK provides standardized, auditable abstractions over third-party grid carbon telemetry. The Green Software Foundation encourages industry developers to participate in open-source projects, standardizing compute hardware energy telemetry, and joining events such as Carbon Hack 2022.