Building an AWS EC2 Carbon Emissions Dataset
summary
Bottom-up EC2 carbon dataset estimates actual power and manufacture emissions: RAPL counters accessed via turbostat on Intel bare metal while load is driven with stress-ng (tooling published as turbostress), yielding CPU+memory peaks above 900 W on a c5 metal. Bare-metal profiles get split by vCPU ratios to estimate instance consumption, AMD/Graviton machines remain unmeasured, and unmeasured components are covered by an added percentage of CPU TDP. Manufacturing is modeled as ~1 tCO2 for a minimal server with a 4‑year lifetime; calculator and data on engineering.teads.com.
AI-written summary · about
chapters (6)
- 0:00Start
- 1:16Why Did We Do this Build this Data Set for Ec2
- 6:27Build Consumption Profiles
- 10:14How Virtual Machines Are Defined by Aws
- 14:00Naive Model To Estimate the Emissions from Manufacturing
- 17:16Cloud Carbon Footprint Project
related talks
41mWriting Greener Software Even When You Are Stuck On-Prem
36mCarbon-Aware Computing: Measuring and Reducing the Carbon Intensity of Software
26mAccelerating Science, Decelerating Carbon: Sustainable Computing...
28mHow to limit your cloud impact, from bare metal to AI
55mToward a carbon-aware Cloud
49mBuilding a zero-waste compute infrastructure on
57mOptimizing your infrastructure for sustainability
28mMore Data Please: Hands on Green Cloud Experiments