Carbon-Aware Computing: Measuring and Reducing the Carbon Intensity of Software
summary
Built a bare-metal cluster and a carbon-aware dispatcher that queries ElectricityMap/CO2 APIs and a cost API to pick time and location for jobs, handling bursts up to 400,000 cores and a test run that produced 15 million files in three minutes. Uses satellite storage to pre-stage inputs, copies outputs asynchronously, and favors spot instances for cost. Time-shifting yields 12–16% CO2 savings; location-shifting can reduce emissions by up to 90%.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 3:52Software carbon emissions matter
- 7:03Carbon intensity and electricity mix
- 12:47Time and location shifting
- 18:12Carbon-aware system with Vestas
- 21:23Measuring SCI
- 26:27Challenges and future
- 30:15Q&A
related talks
26mAccelerating Science, Decelerating Carbon: Sustainable Computing...
42mBuilding a green supercomputer in the cloud
55mToward a carbon-aware Cloud
51mTowards a Carbon-Aware Cloud
41mWriting Greener Software Even When You Are Stuck On-Prem
41mToward a carbon-aware cloud
46mHow to Reduce Your Carbon Footprint with a Scheduled Autoscaling Calendar
49mArchitecting for Sustainability: A Software Architects Guide to Green Software Keynote