Predictable Performance Through Prometheus and Topology Aware Scheduling

Explore how Prometheus and topology-aware scheduling can enhance performance predictability in Kubernetes clusters in this 21-minute conference talk. Learn about the challenges of scheduling performance-sensitive workloads and the limitations of the default Kubernetes scheduler in handling node topology. Discover how exposing cluster-level topology to the scheduler and utilizing Prometheus to track NUMA topology-related metrics can lead to more intelligent NUMA-aware placement decisions. Gain insights into the Topology Manager in kubelet and its role in aligning topology-based resource allocations. Understand how these enhancements to the default scheduler can result in more effective cluster-wide performance of workloads, particularly for performance-sensitive applications.