What's Up With All The Different Container Runtimes? - Ricardo Aravena, Branch Metrics
summary
Explains why Kubernetes' CRI drives a spike in runtimes and gives concrete tradeoffs: Branch runs containerized APIs at 60–70k requests/sec and uses Mesos for Spark/Flink pipelines, while production uses Docker → containerd → runC. CRI lets kubelet call many backends (CRI allows multiple runtimes), so options split by isolation and speed: cri-o keeps containers across daemon restarts, containerd restarts them, crun shows ~40.7% faster than runC without new network namespaces, and Kata Containers trades slower startup for VM-level isolation.
AI-written summary · about
chapters (12)
- 0:00Introduction to runtimes
- 2:52History and evolution
- 6:39OpenVZ, Mesos, and LXC
- 8:57Rocket container runtime
- 10:49Docker and CRI overview
- 13:22Implementing CRI
- 17:20CRI-O and alternative tools
- 19:06Performance benchmarks
- 20:25Kata containers and security
- 24:56Unikernels and summary
- 25:42Runtime selection guide
- 27:39The future of containers
related talks
42mCRI Runtimes Deep Dive: Who's Running My Kubernetes Pod!?
22mRight Abstraction
43mContainers Change Everything
31mReverse Engineering Cloud Native: Interoperability and Community
05mRust Here, Rust There, Rust Everywhere! How a Crab Conquers the Cl...
34mPanel: Improving and Managing Kubernetes at Scale
25mReverse Engineering Cloud Native: Interoperability and Community
46mOk, so I have all these Containers, what now?