Beyond CAS: Why the world needs another Kubernetes Cluster Autoscaler

on YouTube ↗32m

" summary

Built a IBM Cloud Karpenter provider and benchmarked it against Cluster Autoscaler across EKS and IKS using Terraform, Ansible, kube-burner, Prometheus and cloud billing APIs; experiments run four clusters (3 initial nodes, max 10), 1‑hour tests with a 20‑minute spike. Implementation faces IBM API sprawl and custom CRD work. Results show 33% fewer nodes on average, about 50% faster provisioning latency, and cost savings up to 33% on AWS and ~20–23% on IKS.

AI-written summary · about

" chapters (11)

1. 0:00Research project intro
2. 1:06Why autoscaling matters
3. 4:01How Cluster Autoscaler works
4. 6:52Introducing Carpenter
5. 9:44Carpenter on IBM Cloud
6. 11:31Benchmarking setup
7. 13:59Automation with Ansible
8. 15:50Workload simulation with Kube-burner
9. 17:14Results and savings
10. 18:40Does the world need it?
11. 19:45Q&A

" related talks

• 39mCluster API & Cluster AutoscalerErivaldo Lopes · KCD Lima, Perú 2024
• 21mWhy should you bother about cluster autoscaling- KarpenterRaja Ganesan · Conf42 Cloud Native 2022
• 27mHarness the power of Karpenter to scale, optimize & upgrade K8sChakkree Tipsupa · Conf42 Kube Native 2024
• 22mKarpenter Efficient scaling of Kubernetes clustersViktor Velmich · AWS Summit Tel Aviv 2023
• 45mYour K8s Infinity Gaunlet: The Marvel of KEDA and Cluster-API for Infinite ScaleScott Rosenberg · Open Source Summit Europe 2024
• 27mScaling K8s Clusters Without Losing Your Mind or MoneyYael Grossman · Conf42 Cloud Native 2023
• 32mRun Cluster Autoscaler on Your Own Cloud; or, Why Do We Need Cluster API?Aleksandra Malinowska · KubeCon + CloudNativeCon China 2018
• 22mUse Karpenter to scale, optimize & upgrade K8sChakkree Tipsupa · SREday London 2024