Abstract/Summary

The environmental impact of research computing is increasingly a topic of concern for researchers. One of the main contributors of compute-related greenhouse gas (GHG) emissions is the production of electricity to power digital research infrastructures (DRIs). The carbon footprint per unit of electricity consumption (called carbon intensity) depends mostly on the grid’s energy mix, i.e., the share of renewable vs. high-carbon production methods, which varies greatly with time and location. Here, we describe the Climate Aware Task Scheduler (CATS), which lets researchers schedule computing when low-carbon electricity is available. CATS leverages carbon intensity forecasts, e.g., from power grid operators, to find the best window of time to run a particular job with minimal climate cost. CATS also provides an assessment of the carbon savings due to delaying compute as opposed to running it immediately. We demonstrate how the tool benefits researchers in the UK using forecasts of the National Grid’s regional carbon intensity.