CL-R06

Time and region shifting under physical constraints

The lowest-carbon location is not always the correct place to run compute.

Abstract. This programme treats workload placement as a constrained optimization problem: first define the feasible set from service and infrastructure conditions, then compare carbon, water, grid stress and cost without hiding the trade-offs.

Methodology Evidence schema

Long-term research questionWhere and when should flexible compute run after physical constraints?

Active projectCL-P05 / Constraint-aware Compute Routing

Current responsibilityYan Zhang / AI & Data Engineering

Next public outputResearch note + decision-record format

Reference electricity intensity: 79–952 gCO₂e/kWh. Location can radically change operational carbon, but the lowest-carbon region is not automatically feasible once water, grid, latency and sovereignty are included.

Workload scheduling can change carbon outcomes, but no universal reduction factor exists.

Published studies report material reductions under specific workloads, regions and control assumptions. Critical Load treats those results as evidence about opportunity, not as a transferable constant.

41.2%UP TO / GREEN

Cluster-wide carbon-footprint reduction reported in the GREEN evaluation.

12%PEAK POWER

Peak-power reduction reported alongside the GREEN carbon result.

14.5%CARBON / DC-CFR

Carbon reduction reported by the DC-CFR multi-agent control study.

14.4%ENERGY / DC-CFR

Energy reduction reported in the same one-year multi-region evaluation.

Xu et al., GREEN, USENIX NSDI 2025 ↗; Sarkar et al., DC-CFR, AAAI 2024 ↗. Results remain bounded by each study's workloads, regions and control assumptions.

Critical Load research questionWhat remains feasible after latency, sovereignty, capacity, grid reliability, water and local constraints are applied before carbon optimization?

Reproducible control research needs a shared environment, not only headline percentages.

DCRL-Green provides a modular, configurable multi-agent environment for testing how workload scheduling, cooling and battery control interact. It makes assumptions inspectable and gives the programme a concrete base for scenario replication and extension.

Sarkar et al. / Hewlett Packard Enterprise / AAAI 2024 / DOI 10.1609/aaai.v38i21.30580

Sustainability of Data Center Digital Twins with Reinforcement Learning

The environment can support controlled comparisons of carbon, energy and cost strategies before place-specific grid, water, capacity and service constraints are added.

Read the paper ↗Open the repository ↗

Environmental optimization comes after technical feasibility.

Service

Deadline, latency, sovereignty, model and GPU availability.

Capacity

Power, grid headroom, cluster availability and reliability.

Context

Carbon, water, cooling and local environmental state.

Pareto set

Keep trade-offs visible instead of compressing them into one opaque score.

Record

Publish the chosen option and the rejected counterfactuals.

Feasible setF = {r,t | latency, sovereignty, capacity, reliability, deadline satisfied}

Decision objectivemin_r,t∈F [carbon, water constraint, grid stress, cost]

The selected location should be explainable.

Candidate	Carbon	Water	Grid	Service	Decision
Region A / now	High	Medium	Constrained	Pass	Reject
Region B / +4h	Low	High	Available	Pass	Review water
Region C / +2h	Low–medium	Low	Available	Pass	Selected

Illustrative decision-record format. Environmental states are placeholders for the final live evidence inputs, not claims about specific real regions.