LINE's Journey; Road to 4 Million Cores in t...
summary
LINE’s Verda private cloud grows to 4 million cores (150% YoY) using OpenStack and custom work to turn infra from slow requests into API-driven on‑demand resources. They implement a Nova compute driver (not Ironic) to expose bare‑metal “parameter” servers via the unified OpenStack API, add host‑aggregate stock assignment, HA group placement, and GitHub + ArgoCD CI/CD for hardware provisioning. Results: middleware becomes managed, provisioning drops from two weeks to ~10 minutes, and tooling works with standard APIs.
AI-written summary · about
chapters (18)
- 0:00Intro
- 1:26Verda Private Cloud
- 1:58Verda & LINE infra scale
- 2:54Service stack in the Verda
- 4:31Start up period
- 8:44Expansion period
- 13:56New infra period
- 19:32Lesson learned from the 7 years journey
- 21:55Background
- 23:24Requirements for baremetal server management
- 25:21What can be done to archive requirements
- 27:28What is baremetal compute driver
- 28:38New architecture
- 31:24Verda Dashboard - VM / Baremetal server management
- 32:32Deep dive to features
- 32:57Stock management
- 35:43HA Group support with failure domain
- 38:38Deployment procedure
related talks
19m10 years of OpenStack at CERN (From 0 to 300k cores)
20mCERN OpenStack User Story
31mIs OpenStack Still Needed in 2022?
30mOpenStack Works, so now what?
15mThe CERN Cloud Infrastructure: 5 years in 15 minutes
18mOpenStack Playing for an Open World
01h 50mKeynote
52mLet’s build a private cloud – how hard can it be?