DECICE Enhances AI Scheduler and Digital Twin for Smarter, Greener Computing

The DECICE project has introduced important improvements to two of its key components — the AI Scheduler and the Digital Twin — further advancing intelligent, adaptive, and energy-efficient scheduling across the compute continuum.

The AI Scheduler now incorporates new predictive patterns for workload allocation. By analyzing historical trends alongside real-time system data, the scheduler can forecast resource demands more accurately, resulting in improved throughput, reduced latency, and a more balanced distribution of workloads.

The Digital Twin has also been upgraded with node-level power consumption metrics, enabling it to monitor energy usage for individual nodes. This enhancement supports energy-aware and carbon-aware scheduling by providing the AI Scheduler with precise data on power consumption. In addition, both the carbon intensity prediction model and the anomaly detection model have been tuned for higher accuracy and faster responsiveness, allowing for better prediction of environmental impact and earlier detection of irregular system behavior.

With access to more accurate and granular system insights, the AI Scheduler can now optimize workload placement not only for performance but also for energy efficiency and environmental impact. For example, it can prioritize running workloads on nodes with lower power usage or reduced carbon intensity, or proactively reroute jobs away from nodes showing early signs of anomalies.

These improvements strengthen DECICE’s abilities. By combining predictive intelligence with sustainability-focused decision-making, DECICE moves a step closer to achieving high-performance computing that is both powerful and environmentally responsible.

Author(s): Felix Stein, University of Göttingen

Key words: # AI Scheduler #Digital Twin #Energy Efficiency #Predictive Intelligence #Carbon-Aware Computing