Research & Papers

New CWM Algorithm Slashes Datacenter Carbon by 42% Under Deadlines

Researchers achieve 42% median carbon reduction with novel scheduler CWM.

Deep Dive

As datacenters scale, their carbon footprint becomes critical. A new preprint from Dominik Schweisgut (HU Berlin) and co-authors tackles the NP-hard problem of scheduling interdependent workflow tasks under deadlines while minimizing emissions. Their algorithm, CWM (Carbon-Aware Mapping and Scheduling), integrates dynamic programming with efficient heuristics to exploit both renewable energy availability and hardware heterogeneity. The authors prove the problem admits no constant-factor approximation even for single processors, motivating their practical approach.

Evaluated against the prior state-of-the-art CaWoSched, CWM delivers a median 42% reduction in carbon cost when the deadline is twice the makespan of a carbon-agnostic baseline. Notably, CaWoSched itself already beats the baseline by 36%. The 29-page paper, with 11 figures, will be presented at Euro-Par'26. For cloud operators and green computing advocates, CWM offers a concrete path to cut emissions without sacrificing performance guarantees.

Key Points
  • CWM algorithm achieves 42% median carbon cost reduction over CaWoSched for deadline-constrained workflows.
  • Problem is NP-hard with no constant-factor approximation, even for uni-processor systems.
  • Combines dynamic programming with heuristics to exploit renewable energy and heterogeneous infrastructure.

Why It Matters

Practical algorithm for datacenters to cut emissions by 42% without missing deadlines, advancing green cloud computing.