Real-time application of bilevel geographic load shifting using locational marginal carbon intensities

Determine a computationally tractable and operationally viable real-time implementation strategy for the bilevel geographic load shifting scheme that optimizes job placement based on locational marginal carbon intensities, so that it can be applied to continuously arriving workloads in production data center environments.

Background

Lindberg et al. (2022) propose a bilevel program for geographic load shifting that leverages locational marginal carbon intensities to reduce emissions. While the approach shows promise, the paper notes practical hurdles: marginal carbon intensity data is not readily available, and crucially, the pathway for real-time operation is unclear. Data center schedulers must make thousands of placement decisions per second, so any method requires a clear, scalable real-time application strategy.

This unresolved question directly impacts whether such carbon-aware optimization schemes can be deployed in operational data center fleets that handle stochastic, continuously arriving workloads. Clarifying the real-time implementation would bridge the gap between planning formulations and production-grade scheduling under stringent latency and capacity constraints.