Exploiting Temporal Diversity of Water Efficiency to Make Data Center Less "Thirsty"

Data centers, which include both cyber (e.g., servers) and physical (e.g., cooling units) assets, are notorious for their energy consumption and carbon footprint. Nonetheless, a less-known fact about data centers is that they are extremely “thirsty” (for cooling), consuming millions of gallons water each day and raising serious concerns amid extended droughts. To curtail the surging water footprint, we adopt a holistic cyber-physical approach and incorporate the inherent physical characteristic of data center— time-varying water efficiency — into server provisioning and workload management. Specifically, we propose an online batch job scheduling algorithm, called WACE (minimization of WAter, Carbon and Electricity cost), which dynamically adjusts server provisioning to reduce the water consumption by deferring delay-tolerant batch jobs to water-efficient time periods. We demonstrate the effectiveness of WACE via trace-based simulations, showing that WACE reduces 27% water consumption compared to state-of-the-art scheduling algorithms.