GreenPASS optimization
Collaborative workload shifting framework designed to reduce data center carbon emissions without increasing user costs.
September 2025 – Present
Collaborators: Zihan Pan (PhD Candidate), Mohammad Shahrad (Assistant Professor), Geoffrey Bian (UREP Affiliate) Affiliation: UBC Cloud Infrastructure Lab (CIRRUS), University of British Columbia
Research Summary
The core of our evaluation utilizes a simulation-based methodology scaled to cluster-level production traces from the Alibaba v2020 GPU dataset. By simulating workloads across four North American regions (Virginia, California, Texas, and Ontario), we analyzed the feasibility of shifting ~70,000 jobs based on Carbon Intensity (CI) and Locational Marginal Pricing (LMP).
Key findings from our research include:
- Cost-Neutral Sustainability: GreenPASS can achieve significant carbon reduction while eliminating the typical 4% to 10% cost overhead associated with user-managed shifting.
- Data Access Synergy: Our Duplication & Reuse policy demonstrates that co-locating data with shifted workloads is essential; reusing data across recurring jobs further reduces shifting costs and maximizes carbon savings.
- Robustness Across Time: Simulations spanning five years (2020–2024) confirm that carbon reduction remains consistent despite seasonal and annual variations in grid intensity and electricity prices.
Publication Status
The paper detailing this framework and its evaluation is currently pending approval for the 17th ACM International Conference on Future and Sustainable Energy Systems (ACM e-Energy 2026).