Burstable workloads with OKD on ARM
summary
Uses OpenShift Ansible and an AWS CloudFormation template to build OKD on ARM EC2, spinning 3-node cluster (1 master + 2 workers) from a fedora/centos image on a1.4xlarge instances; master performance on ARM is slow so Python2 and RPM installs are required for some components. Demonstrates autoscaling with auto scaling groups and scalar.py plus life cycle hooks to drain and register nodes, claims ~45% cost savings for burstable workloads, and shows the setup mostly works though transient failures and rate limits occur.
AI-written summary · about
chapters (7)
- 0:00ARM and OKD intro
- 3:24Burstable workloads on ARM
- 13:02AWS console setup
- 21:11VPC and nodes via CloudFormation
- 33:04Cluster instance config
- 37:14Auto-scaling for bursts
- 40:03Q&A on ARM challenges
related talks
44mThe Road to OKD4: Operators, Fedora CoreOS and K8S
36mBare metal Kubernetes on mixed x64 and ARM64
32mHow we build OKD (Kubernetes distro) using Tekton
36mOKD Working Group: State of OKD4/FCOS
38mRunning Multi-Arch Builds & Tests in OpenShift CI/CD
10mLightning Talk: Does Cloud Elasticity Pay off for HPC Workloads?
44mPerformance and Scalability Tuning Kubernetes for OpenShift and Docker
42mDrupal, Kubernetes, Production