Do Not Disturb Mode: Fine-Tuning Resources for Latency Sensitive Workloads

on YouTube ↗21m

" summary

Kubernetes gains node-local class-based QoS via cri-o 1.22 enabling Intel RDT for cache and memory bandwidth and a block I/O class file for cgroup-level scheduling. Static CPU manager pins cores, RDT assigns cache partitions and memory bandwidth limits, and block I/O classes set weights or hard throttles (demo shows md5sum reader limited to 5 MB/s via a slow reader class and kernel read bps) and boosting a critical service with block IO weight 400 reduces latency on shared nodes while running successfully on a VM using per-node configs.

AI-written summary · about

" chapters (4)

1. 0:00Start
2. 13:56Class-based Qos
3. 16:43Food for thought: QoS future
4. 20:09Tuning workloads... done.

" related talks

• 36mClass Resources: Kubernetes’ Fastest Way Of Shushing Noisy NeighborsMarkus Lehtonen, Peter Hunt · KubeCon + CloudNativeCon North America 2022
• 36mOptimizing Kubernetes for Low-Latency and Real-Time ContainersAntti Kervinen, Feruzjon Muyassarov, Alexander Kanevskiy · ContainerDays Hamburg 2026
• 34mUnlocking Performance: Topology-Aware CPU Scheduling W...Praveen Krishna, Marlow Warnicke · CNCF Maintainer Summit North America 2025
• 23mThis is The Way: A Crash Course on the Intricacies of Managing CPUs...Marlow Weston, Swati Sehgal · KubeCon + CloudNativeCon Europe 2022
• 36mMitigating Noisy Neighbours: Advanced Container Resource ManagementAlexander Kanevskiy · KubeCon + CloudNativeCon North America 2019
• 24mMaximizing Workload's Performance With Smarter RuntimesKrisztian Litkey, Alexander Kanevskiy · KubeCon + CloudNativeCon Europe 2021
• 32mCoordinate Workloads Colocation: QoS-Oriented Scheduling Enhancement on K8sTao Li, Zuowei Zhang · Kubernetes Batch + HPC Day North America 2022
• 27mWhat’s in the Box? Resource Management in KubernetesLouise Daly, Ivan Coughlan · KubeCon + CloudNativeCon Europe 2018