Using Unikernels to deploy real-world apps
summary
Genesio runs production experiments using Firecracker to boot OSv unikernels that bundle runtime with a tiny kernel, taking snapshots and keeping a snapshot warm pool to cut cold starts. Cold round trips measure ~70 ms (Genesio unikernel/Linux) versus ~300 ms for AWS Lambda; warm requests fall to ~30 ms versus Lambda’s ~60 ms. User code mounts as a separate disk so base images can be upgraded without redeploying apps, and Firecracker Jailer plus vmm-level block sharing enforce isolation while reducing memory and storage use.
AI-written summary · about
chapters (12)
- 0:00Unikernels introduction
- 1:27What unikernels are
- 4:37Genesio’s orchestrator
- 7:05Optimizing cold starts
- 9:15Upgrading unikernels
- 10:40Benchmark vs AWS Lambda
- 13:54Future multi-language plans
- 14:18Q&A: Storage
- 16:28Q&A: New features
- 17:53Q&A: Snapshots strategy
- 20:48Q&A: Scalability overhead
- 23:19Closing remarks
related talks
38mUnikernels: The Next Evolution in Cloud Infrastructure
34mLook ma, no OS! Unikernels and their applications
17mUnikernel based deployments
44mUnikernels and Hyper Elastic Clouds
23mCrack the FaaS Cold Start and Scalability Bottleneck
27mServerless MicroVMs for Containerized Applications
24mCut your Cloud Computing Costs by Half with Unikraft
10mLightning Talk: Millisecond Scale-to-Zero and the Quest to Never Pay for Idle Again