KubeCon + CloudNativeCon Europe 2025

Multi-cluster AI infrastructures have become the norm due to factors such as resource availability, platform scale, high availability, and business resource pool consolidation. However, managing diverse workloads across heterogeneous clusters can be challenging. In this talk, we will share our experiences and lessons learned from deploying Karmada and Volcano in real-world multi-cluster AI environments. We will delve into specific Day2 optimization techniques, including:
1) Configuring scheduling strategies to balance resource utilization and workload priorities.
2) Customizing workload management to accommodate diverse AI workloads with varying requirements.
3) Leveraging topology-aware scheduling to improve the efficiency of AI training and inference tasks.

By sharing concrete examples and results, we will demonstrate how to effectively optimize multi-cluster AI infrastructures to achieve better performance, cost efficiency, and scalability.