Honey, I Blew Up the Cloud
summary
Bloomberg rapidly scales single-cell OpenStack clusters to tens of thousands of VMs with Ceph-backed volumes and thousands of security groups, and describes concrete fixes: rewiring control-plane topology to a shared-nothing model, distributing the metadata service to avoid DB query hot spots, patching Neutron ORM to eliminate API brownouts, and automating lifecycle ops for zero-downtime upgrades. Metrics show the cluster stabilizes after these changes and most scaling pain points are resolved.
AI-written summary · about
chapters (24)
- 0:00Start
- 0:15About us
- 1:12Agenda
- 3:01Use-cases
- 3:44Geographically Distributed Edge Architecture
- 6:58Low Latency Stretched Architecture
- 10:43Compute
- 12:54Networking Infrastructure
- 15:06Routed Provider Networks
- 16:08Which Neutron network to use?
- 17:54Glance Images at the Edge
- 21:10Persistent Volumes at the Edge
- 22:38Container Image Registry
- 24:34Red Hat OpenStack Edge Roadmap
- 26:42Lessons Learned from the Field
- 33:57The Slope of Enlightenment
- 34:27Superior economic advantages
- 34:37Highly performant infrastructure
- 37:04Introducing
- 37:59Infrastructure for infinite possibilites
- 38:32Get started with OpenStack today
- 40:25About me...
- 44:06Designing for scale
- 57:37Cache stampedes