Bloomberg's Journey to Manage Multi-Cluster Training Applications with Karmada

Learn how Bloomberg tackled the challenges of managing a large-scale, multi-cluster GPU environment for AI model training in this technical conference talk. Explore the implementation of an on-premises Data Science Platform built with cloud-native software and running on Kubernetes across multiple data centers. Discover solutions for intelligent GPU workload scheduling, resource utilization optimization, and operational cost reduction through the integration of Karmada. Gain insights into advanced techniques including the use of Karmada's resource interpreter for Custom Resource Definition support, the construction of a highly available control plane, and the development of a streamlined training job submission interface. Follow Bloomberg's Data Science Platform team's journey as they collaborate with the Karmada community to solve complex multi-cluster batch job management challenges across diverse GPU environments.