From Screen to Pods: Bootstrapping a Cloud Agnostic System using Kubernetes [I]
summary
Algorithmia migrates from AWS VM-based deployment to a Kubernetes-first platform, running across multiple clouds and on-prem in under an hour, reducing required servers from 12 to 6. The system uses Docker images, CNI (weave), nginx ingress, Prometheus and Elasticsearch, handles CPU and GPU nodes, exposes GPUs to containers, and learns around AWS ELB and EBS quirks. Production runs with 100s of pods and 10,000s of containers created daily, and the migration succeeds.
AI-written summary · about
chapters (35)
- 0:00Intro
- 0:12Agenda
- 0:34About me
- 0:55What we do
- 1:44Filters
- 2:22Outline
- 2:45Web Application Diagram
- 4:23Limitations
- 5:10Public Marketplace
- 6:00Barriers
- 6:40Benefits
- 7:44Caveats
- 9:42Storage drivers
- 10:39Upgrading kernel packages
- 11:20EBS volume performance
- 12:43Questions
- 13:00Why Kubernetes
- 13:42Kubernetes Services
- 14:51Why we dont use Kubernetes
- 16:35GPU and Device Management
- 19:17Prometheus Cortana
- 19:49What went wrong
- 21:44New concepts
- 22:24Technical challenges
- 23:15Classic load balancers
- 25:10Cluster Groups
- 25:26Network topology
- 26:10Moving to multiple clouds
- 26:28Considerations
- 27:06Provisioning
- 27:25Azure equivalents
- 29:09Issues with Load Balancers
- 30:24Azure Traffic
- 31:04Summary
- 31:28Looking Forward
related talks
35mKubernetes journey
01h 25mTutorial: Migration 101: From VMs to Kubernetes
35mNavigating the Kubernetes Odyssey: Lessons from Early Adoption and Sustained...
46mMigrating from VMs to K8s
32mMigrating 2 million CPU cores to Kubernetes
29mFrom EC2 to Kubernetes Takeaways from a Brownfield Migration
42m18 Months Before the Mast
25mBreaking Free From the Cloud: Banking on Self-Hosted Kubernetes