Oleksiy’s talk evolves around how to mitigate issues “carbon aware scheduling” when you cannot stop and shift workloads:
- Techniques are power capping, frequency capping, utilization capping etc.
- Systems are typically non linear with steps from 80→100% performance costing the most per step
- Different techniques (DVFS, RAPL powercapping, CGroup Utilization capping) deliver different output characteristics.
- A mix of all modes can lead 2x gains of performance drop. (aka 5% performance drop and 10% cost savings)
If you liked the Sneak Peak you can find the full slides on the ecoCompute website: https://www.eco-compute.io/files/slides_2025/01_Thursday/01_SoHa/10_Kozlov_Onepedal.pdf