Evaluating Carbon-Aware Scheduling for Data Centers: A Grid Simulation Framework

Abstract

The proliferation of power-hungry data centers, driven by the rapid emergence of artificial intelligence (AI), places significant strain on existing electricity grids, while also raising environmental concerns related to the carbon emissions associated with new electricity generation. By shifting workloads across space and across time, data centers can potentially reduce the total needed generation capacity, as well as the carbon emissions and costs associated with the data centers’ electricity needs. This thesis utilizes the Vatic model of the ERCOT electricity grid in Texas, adapted to account for the expected additional electricity needs of future data centers, to examine the extent to which carbon-aware scheduling (CAS) can improve outcomes. A CAS strategy that jointly targets emissions reductions and cost reductions outperforms two other CAS strategies, one that shifts workload to periods when electricity generation has lower carbon intensity, and another that shifts workload to periods of higher renewables generation.