Practice of Fine-grained Cgroups Resources Scheduling in Kubernetes
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Thursday, 2020-11-19· at 17:40· Intrado Virtual Event Platform,·
summary
Alibaba runs scheduling across hundreds of thousands of nodes and millions of containers and describes a Kubernetes-based system that does dynamic cgroup reconfiguration and NUMA-aware CPU pinning without restarting Pods. It uses a Scheduler Framework plugin set plus a cgroup controller and node-side agent, creates cgroup CRs and executor Jobs to apply limits, supports policy CRDs to rebalance resources, and reports real results: ~20% lower Java latency, 20–30% QPS gain, ~10% faster video transcode, and +13% insert / +10% select TPS.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:40Agenda
- 1:58Cgroup Resource Management
- 3:14What is Cgroup Controller?
- 5:13Cgroup Controller Implementation
- 6:57Cgroup Controller in Practice
- 7:53Smart Cgroup Controller
- 10:03Cgroup Policy in Practice
- 12:25Why CPU scheduling is needed ?
- 13:41Architecture
- 15:15Scheduling Framework Plugins For CPU Scheduling
- 16:29Scheduling algorithm
- 18:19Statistical results
- 19:28Descheduler
- 21:01scheduler-plugins
- 21:49Future works
related talks
36mCapacity Scheduling for Elastic Resource Sharing in Kubernetes
34mUnlocking Performance: Topology-Aware CPU Scheduling W...
24mKubernetes Batch Processing at Scale - A Scheduling Perspective
38mDynamic Pod Resource Boundary Adjustment in Web Scale Clusters
29mScale Your Batch / Big Data / AI Workloads Beyond the K8s...
37mCo-Location of CPU and GPU Workloads with High Resource Efficiency
07mHow to Handle Fair Scheduling in a Private Academic K8s infrastructure
25mKube-scheduler-evaluator: Evaluating Kubernetes Schedulers at Hyperscaler…