TR-2020-09
Using Carbon-awareness to Reduce the Cloud’s Environmental Impact
Chaojie (Sam) Zhang; Andrew A Chien. 4 November, 2020.
Communicated by Andrew Chien.
Abstract
Carbon emissions and power are a challenge for the growth of cloud computing. We consider a carbon-aware approach to acquiring power that caps emissions hourly, adapting to the carbon content of the power grid. The objective of this approach is to reduce the cloud’s carbon footprint.Because this approach produces variable capacity, we assess performance on production cloud workloads (Microsoft, Google), exploring several dimensions of variation, dynamic range, structure, and change frequency. The robust high performance shows that carbon-aware policies need not damage cloud efficiency. With foresight and scheduler improvement, potential job impact is minimized under extreme dynamic range.
To demonstrate the potential benefits, we consider a case study with a cloud datacenter in Germany. Our results show that the constant hourly emission approach can reduce carbon footprint by up to 19% while maintaining system performance. An additional benefit is lower power cost.
Original Document
The original document is available in PDF (uploaded 4 November, 2020 by Andrew Chien).