Sustainable Computing
Carbon-aware scheduling, cloud decarbonization, and the tradeoffs between carbon, energy, and cost
The rapid growth of cloud and edge computing is driving a sharp rise in data center energy consumption and carbon emissions. My research develops systems and algorithms that make cloud, edge, and residential computing more carbon-efficient by shifting flexible workloads to times and places of low-carbon electricity, while carefully managing the resulting tradeoffs with cost, energy, and performance.
Carbon-Aware Scheduling and Scaling
A key insight behind carbon-aware computing is that the carbon intensity of the electrical grid varies significantly over time and space. By deferring or migrating flexible workloads to windows of cleaner electricity, we can reduce operational carbon emissions with little or no impact on job completion time.
CarbonScaler (Hanafy et al., 2024) leverages cloud workload elasticity (scaling workloads up or down) to run computation during low-carbon periods. Rather than treating carbon as a separate objective, CarbonScaler integrates carbon signals directly into the auto-scaling loop, achieving Best Student Paper at ACM SIGMETRICS 2024.
CarbonFlex (Hanafy et al., 2025) extends this to cluster-level provisioning, enabling carbon-aware scheduling across heterogeneous workloads and multi-tenant cloud environments.
Carbon–Cost–Energy Tradeoffs
Reducing carbon is not free, as shifting computation in time or space can increase monetary cost or energy use. Understanding and managing these tradeoffs is central to practical decarbonization.
GAIA (Hanafy et al., 2024) formalizes the cost of carbon reduction in the cloud and develops optimization strategies that minimize carbon subject to a cost budget. Published at ASPLOS 2024, GAIA shows that significant carbon reductions are achievable with modest cost increases.
Untangling the Carbon–Cost Tradeoffs (Hanafy et al., 2026) further examines the “stampede effect” when many workloads simultaneously shift to the same low-carbon window, causing resource contention that undermines both cost and carbon goals.
The War of the Efficiencies (Hanafy et al., 2024) shows that carbon-optimal and energy-optimal schedules are often in conflict, where optimizing one can significantly worsen the other, with important implications for how data centers should report and optimize their environmental impact.
Spatial Shifting and Edge Decarbonization
Beyond temporal shifting, workloads can also be migrated spatially by routing computation to geographic regions with cleaner grids.
CarbonEdge (Wu et al., 2025) exploits mesoscale spatial variations in carbon intensity across edge computing sites to reduce the carbon footprint of edge inference and data processing.
CDN-Shifter (Murillo et al., 2024) applies spatial shifting to content delivery networks, redistributing traffic across CDN nodes to favor low-carbon regions, reducing emissions at internet scale.
Beyond the Data Center
Carbon-aware principles extend to residential energy and other societal infrastructure.
GreenThrift (Bovornkeeratiroj et al., 2025) optimizes scheduling of flexible residential loads (EVs, HVAC, appliances) to minimize both carbon emissions and electricity costs, bridging the gap between grid-scale decarbonization and individual households.
A broader vision for Computational Decarbonization (Irwin et al., 2025) articulates how computing systems, from data centers to mobile devices, can be redesigned to actively reduce the carbon footprint of societal infrastructure.
References
2026
2025
-
Arxiv
-
JCSSGreenThrift: Optimizing Carbon and Cost for Flexible Residential LoadsACM Journal on Computing and Sustainable Societies, Mar 2025Presented in COMPASS 2025
