SoftBank & Red Hat Optimize Power for AI-RAN Data Centers
Barcelona, Spain / Tokyo, Japan - SoftBank Corporation and Red Hat, Inc. have jointly developed a solution to optimize power consumption in data centers operating vRAN (virtualized Radio Access Network) and AI applications for AI-RAN.
The solution has been added to the AITRAS orchestrator, enabling dynamic resource allocation based on power usage metrics for AI applications.
The development addresses high power requirements for stably operating vRAN and AI applications in AITRAS data centers.
This comes amid growing initiatives to reduce greenhouse gas emissions through renewable energy adoption and carbon intensity evaluation.
Additionally, the concentration of data centers in metropolitan areas has created demands for enhanced resilience and distributed power loads.
Red Hat has customized the Kepler project's power monitoring functionality and made it available as a Red Hat OpenShift feature.
Kepler operates in Kubernetes environments, collecting server and application power information from clusters throughout data centers.
It stores this data as metrics that can be provided to software such as the AITRAS orchestrator.
Kepler also delivers power information calculated based on GPU chip and MIG (Multi Instances GPU) usage rates.
SoftBank incorporated Kepler software into the AITRAS virtualization infrastructure based on Red Hat OpenShift.
This enables the orchestrator to consider power-related metrics when optimizing resource allocation, improving power consumption efficiency.
The AITRAS orchestrator can now perform optimization functions, including:
- Placing applications in clusters with lower power consumption in multi-cluster environments
- Setting maximum power consumption limits per cluster
- Configuring parameters such as "available power from renewable energy" and carbon intensity values as placement conditions
The solution will be exhibited at Red Hat's booth during MWC Barcelona 2025.