The rapid proliferation of AI across scientific and industrial domains has fundamentally challenged traditional compute infrastructure paradigms.
High-Performance Computing centers, long governed by battle-tested schedulers such as Slurm, PBS Pro, and IBM LSF, now face mounting pressure to accommodate containerized AI workloads that demand the elasticity and ecosystem richness of Kubernetes.
This talk explores the emerging pattern of hybrid orchestration — architectures that navigate the coexistence of traditional HPC schedulers and Kubernetes rather than forcing a choice between them. We examine what each orchestrator excels at, how different workload types naturally gravitate toward different schedulers, and the real-world complexity of managing resources across both worlds.
From there, we dive into the practical challenges organizations encounter at the seams: data locality, storage interoperability, security boundaries, and operational overhead.
The talk concludes with current and emerging approaches to bridging the gap, an industry outlook, and actionable guidance for teams navigating this transition.
