When Your Small Cloud Stopped Being Small
summary
Covers problems of organic growth of cloud networks at very large scale—how L2 domains spanning tens of thousands of machines arise and how physical and virtual networks behave. Describes vendor network-equipment quirks, typical customer requirements that drive network-service evolution, and Timeweb Cloud’s experience scaling solutions (physical VTEP vs Linux kernel, L3 direct connect vs L2).
AI-written summary · about
chapters (21)
- 0:00Introduction
- 1:10Building new cloud network
- 2:14Evolution of flat network
- 3:21L2 domain growth issues
- 5:09Large segments and tables
- 7:15Why L2 plane is inefficient
- 9:00CPU load on network gear
- 10:05Separating network functions
- 13:37Stabilizing current infra
- 14:18Managing tables
- 16:35MAC address overload
- 17:18Broadcast storm issues
- 18:22Solving L2 network problems
- 19:48BGP on hypervisors
- 21:14Hardware vs software
- 22:49Direct client connections
- 24:42Overlay services
- 26:33SDN and Linux evolution
- 28:02When to switch architecture
- 28:46Switch scaling problems
- 30:15Q&A
related talks
38mHow can a network be "virtual"?
33mAnd Then the Clouds Will Surely Start to Dance!
20mKeynote: Driving Network Automation and Cloud Native
46mAWS Networking : Building your Network from 0 to Millions of Clients
16mModern networking: Clever Cloud network groups, logical and physical separation
21mQ&A Session 1 – Luca Niccolini
24mNetwork Visibility in Hybrid and Multi Clouds
40mCloud, Kubernetes, and Service Networking - Taming the Turtles