Let's Wait Awhile: How Temporal Workload Shifting Can Reduce Carbon Emissions in the Cloud

Overview

This paper, published in 2021 in the Proceedings of the 22nd International Middleware Conference (ACM), examines how much carbon can be saved by delaying delay-tolerant data-center workloads to times when the grid is cleaner. It uses 2020 carbon-intensity data for Germany, Great Britain, France and California. The authors characterise shiftable workloads (batch jobs, scheduled and periodic tasks, interruptible jobs) and analyse daily and weekly carbon-intensity patterns per region. Mean 2020 carbon intensity was 311.4 gCO2/kWh in Germany (range 100.7-593.1), 211.9 in Great Britain, 56.3 in France and 279.7 in California.

Weekend carbon intensity is lower than weekdays: Germany 243.7 versus 328.7 gCO2/kWh (25.9% lower), Great Britain 20.7%, France 22.2% and California 6.2% lower. In California, on 44% of days in 2020 the carbon intensity of workloads scheduled at 6 am could be cut by more than 80 gCO2/kWh by shifting, owing to the drop before sunrise. Simulations of two shifting scenarios test the effect of time constraints, scheduling strategy and forecast accuracy.

Savings depend strongly on the regional mix; the largest gains come from relaxing deadlines and exploiting interruptibility, and weekend shifting is attractive in most regions. Shifting delay-tolerant workloads toward weekends can save more than 20% of emissions in most regions, and shifting workloads whose results are not needed until the next working day already reduces emissions by over 5% across all regions. Cited workload studies show batch jobs make up 60% of processing on large clusters, and more than 40% of these run daily.

Key points

1. Mean 2020 carbon intensity: Germany 311.4 gCO2/kWh (range 100.7-593.1), Great Britain 211.9, France 56.3, California 279.7.
2. Weekend carbon intensity is lower: Germany 243.7 vs 328.7 gCO2/kWh (25.9% lower), Great Britain 20.7%, France 22.2%, California 6.2% lower.
3. In California on 44% of days in 2020, the carbon intensity of 6 am workloads could be cut by more than 80 gCO2/kWh by shifting.
4. Shifting delay-tolerant workloads toward weekends can save more than 20% of emissions in most regions.
5. Shifting workloads not needed until the next working day already reduces emissions by over 5% across all regions.
6. Batch jobs make up 60% of processing on large clusters, and more than 40% of these run daily.
7. Largest gains come from relaxing deadlines and exploiting interruptibility; savings depend strongly on the regional grid mix.

Using it in design

• For carbon and sustainability reporting, use hourly (not annual average) grid carbon-intensity profiles per region.
• Offer 'delay-tolerant share' and 'deadline' inputs to estimate shifting savings.
• Show that the benefit depends strongly on the grid mix: high for wind/solar grids such as Germany, low for nuclear/hydro grids such as France.
• Evaluate site selection and scheduling assumptions together.

Who should read it

Data center and cloud engineers doing carbon-aware workload management and sustainability reporting.

Publication details

Authors

Philipp Wiesner; Ilja Behnke; Dominik Scheinert; Kordian Gontarska; Lauritz Thamsen

Publisher

Proceedings of the 22nd International Middleware Conference (ACM)

Year

2021

Document number

arXiv:2110.13234

Pages

260-272

Document language

English

Source type

Scientific

Access

Free to read

DOI

10.1145/3464298.3493399

Licence

arXiv distribution permission — reuse conditions remain with the rights holder

Citation