A road trip into AWS and back to bare metal
summary
A team moves a heavy image pipeline (serving 60 million products and about 2,000 images/sec, ~1,000 CPUs and 1,000 searches/sec) into AWS, learns that spot instances save money but spur complexity, and replaces brittle handcrafted VMs with Ansible provisioning. Performance, networking costs and vendor lock-in push a return to bare metal using PXE/IPMI and modern chassis, while Kubernetes becomes the portability layer and AWS remains a DR/elastic burst option.
AI-written summary · about
chapters (14)
- 0:00Company and project overview
- 2:57Journey into AWS
- 4:44Image delivery use case
- 7:13AWS infrastructure overview
- 10:41Operational challenges
- 14:24Ansible automation
- 15:32Cost and scalability issues
- 17:34Spot instance strategy
- 23:26Vendor lock-in and tooling
- 26:51Ugly architectural patterns
- 30:52Moving back to bare metal
- 34:55Cloud principles on-prem
- 38:22Kubernetes and conclusion
- 41:29Q&A session
related talks
49mHardware to AWS, EC2 to EKS: Two Years, Two Epic Migrations
50mBare Metal DevOps
01h 02mTransforming IT pros to DevOps gurus: Ssecure your new tech stacks
38mAccelerating innovation through AWS Cloud technology
23mA Journey to Automated Infrastructure
30mLeaping from Mainframes to AWS: Technology Time Travel in the Government
31mDevOps in the Cloud: Case Studies of Amazon.com teams and their resilient architectures
38mBuilding Data Pipelines on AWS