Inception: Towards a Nested Cloud Architecture
summary
Proposal builds a nested IaaS called Inception that runs nested KVM VMs inside regular cloud VMs (RVMs) and connects inner VMs (IVMs) with a virtualized layer‑2 network using Open vSwitch + VXLAN. It runs a stock OpenStack unmodified control plane, automates deployment with Chef and produces a working nested cloud in about 20 minute deployment. Prototype shows CPU/memory overhead ~5–10% but notable network/storage bottlenecks; plans use SDN controllers to scale to ~10,000 IVMs per data center and support cross‑provider live migration and floating IPs.
AI-written summary · about
chapters (10)
- 0:00Start
- 5:05Inception
- 7:48Application Scenarios
- 8:39Continuous Integration
- 11:45Inception Vm
- 12:25Network Design
- 15:06Private Network Design
- 16:21Public Network Design
- 21:52How Do You Propagate Multicast
- 22:40Performance Overhead
related talks
41mCloud Computing with KVM
05mFrom Open Source to Real-World Scale
16mInvisinets: Removing Networking from Cloud Networks
01h 21mModern Infrastructure: The Convergence of Network, Compute, and Data
23mBYO Infrastructure: Who needs to run a control plane, anyway?
38mYour Own Little HA Cloud at the Edge (or at Home)
42mEnhancing Hypervisor and Cloud Solutions Using Embedded Linux
44mHow we designed an OpenStack Public Cloud at Scale with Fabric Networking