taylorprewitt / carbon-awareapi

Increasing carbon efficiency and footprint awareness for software applications.

Carbon Aware API

^{Draft, Project In Progress}

Navigation

1. Overview
2. Case Studies
3. API Architecture
4. Methodology
5. Sample User Interface

Terminology

Term	Definition
Carbon Intensity	Carbon emitted per energy unit.
Grid-Based Carbon Intensity	All entities who operate on a shared electrical grid, share a common carbon intensity.
Marginal Carbon Intensity	The carbon intensity of the marginal power plant that supplies power when additional load added to the grid.
Carbon Aware	Adjusted behavior in response to the carbon intensity of consumption.
Carbon Delta	The difference in emissions between carbon aware and unaware actions.
Carbon Counterfactual	The carbon delta had the carbon aware action been different.
Demand Shifting	Selectively changing the time/location of a compute's execution, to a time/location where the energy demands are met by cleaner energy production, resulting in a lower grid-based carbon intensity.
Operational Emissions	Emissions explained by the energy consumption and grid-based carbon intensity measurement during times of operation.
Embodied Emissions	Carbon emissions resultant of creating the hardware, structural systems, maintanence, etc. (e.g. constructing a GPU or datacenter).

Other definitions and methods for the Carbon Awareness found at: Green Software Foundation

Overview

To enable organizations to make smart decisions about their environmental impact and carbon footprint, we have created the Carbon Aware API to minimize the operational emissions of computational workflows. A few key features of the API are:

• Utilizes marginal carbon emission measurements from WattTime to map carbon intensity rates to data centers.
• Generates a retrospective carbon emission analysis
• Provides forecasted demand shifting suggestions
• Supplies regional carbon intensity information over different scopes
• Hosts a green region picker for workspace configuration.

Marginal Carbon Emissions: This grid-responsive metric with finer granularity than average emissions, allows for seasonality/diurnal trends captured in demand shifting (source).

Retrospective Analysis: Time series evaluation to assess the carbon emissions for a given energy profile. Also provides counterfactual analysis to expose the potential emissions of if the run had been shifted.

Demand Shifting Scheduler: Recommends data center and/or time which would yield a less carbon intensive run.

• Temporal: Identifies the window of minimum carbon intensity for a specified run duration within a chosen region from a 24-hour forecast.
• Geographic: Finds the region with the current lowest average carbon intensity for an immediate run of a specified duration. Can filters available regions by available SKU and migration laws for workspaces with protected data.

Regional Carbon Intensity: Provides the carbon intensity for each data center supported by a WattTime-tracked balancing authority. The possible scopes are historic intensities (time series for prior 24-hours, week, and month), real-time marginal intensity, and forecast (mean intensity for upcoming user-defined window).

Green Region Picker: Recommends the a data center to host a workspace based on expressed needs such as carbon efficiency, price, and latency.

For more information, please see the full tool description: Carbon Aware API Details

Case Studies

^{Carbon intensity data for case studies provided by: WattTime}

Carbon Optimized Demand Shifting at Scale

^{September 2021}

Details: Geographically and temporally shifting compute instances at scale would reduce carbon by X%

Results: Organizations can reduce their operational emissions by 48% by geographically shifting and 12% by temporally shifting ML computes.

For more information and detailed results, please see the full case study: Carbon Optimized Demand Shifting at Scale

Carbon Segmented Cloud Computes

^{October 2021}

Details: Execute and progress cloud computes greater than 24 hours in duration during periods of low emission to reduce the carbon footprint by Y%

Results: 48-hour ML runs saw a mean of 0.5% change using standard temporal shifting, but controlling progression based on regional carbon intesity thresholds reduced operational emissions by 10%

For more information and detailed results, please see the full case study: Carbon Segmented Cloud Computes

carbon-awareapi's Issues

Swagger

Implement automated documentation generation for the Swagger, preference is as YAML.

System

Implement system testing on the Carbon Aware API

CI/CD Pipeline

Set up a pipeline for CI/CD testing/deployment

Integration

Implement integration testing on the Carbon Aware API

CI data source modularity

Enable code to be agnostic to the CI data source. Ex: enable WattTime and ElectricityMap, etc

Scalable Framework

Move UI from HTML to JavaScript in React

Impact of API

Build a method for carbon counterfactual tracking of scheduling outcomes

API usage

Build method to track usage of individual endpoints. Total usage and outcome of use.

Session Issues

Address endpoints with session issues and dependencies to prevent errors

Classification

Organize functions/processes and structure based on OSS standards

Readthedocs

Implement automated documentation generation for readme docs.

Tool Validation

evaluate the carbon API and its ability to be used as a smart scheduler for each shifting type

Unit

Implement unit testing on the Carbon Aware API

Hosting CSP agnostic

Expand tool beyond Azure data centers to accept inputs for other CSP datacenters

Style Standards

Ensure all code conforms to PEP8 standards

Refactor Endpoints

Refactor the functions/code base to increase robustness, decrease error response frequency, decrease latency, minimize code