One-cloud — Data Center Management System at Odnoklassniki
summary
One-Cloud is a data-center management system used at Odnoklassniki to control over eight thousand physical servers across multiple data centers. It organizes each machine into specialized roles to provide fault isolation and enable automated infrastructure management, and the talk explains the system’s architecture and operational practices for that automation.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 5:20Low resource utilization problem
- 6:45Containers vs VMs
- 10:00One-cloud architecture
- 11:29Resource limits
- 13:16Hierarchical queues
- 16:15Task classes and usage
- 21:34CPU task prioritization
- 25:56Network traffic prioritization
- 28:52Task isolation limits
- 31:24Container fault tolerance
- 33:07Container failover
- 36:57DNS service discovery
- 38:23BGP traffic routing
- 41:16Datacenter outage response
- 43:04Task priorities and eviction
- 45:09Datacenter fault isolation
- 47:16Operator error protection
- 48:40One-cloud summary
- 49:25Q&A
related talks
35mBehind the Scenes of the Cloud
20mMulti-Cloud Without the Myth: What Actually Works in Practice
15mThe CERN Cloud Infrastructure: 5 years in 15 minutes
47mHow (not) to survive a data center outage
22mJust in Time Infrastructure
33mIntent-based holistic data center management
49mOperation Crossplane: Cloud Command Tactics
24mHow to survive Cloud – An Ops perspective