Running Unikernels in Prod for a Few Years
summary
NanovVMS runs unikernels in production for 3–4 years using a custom kernel called Nanos and tooling that packs ELF binaries into 20 MB image raw disk images, boots them as VMs (not containers), and deploys to Google/Amazon with customers in prod. Measurements show 200% faster on Google and up to 30% faster on Amazon depending on hypervisor. Images are single-purpose with no SSH or users, support x86/arm/risc64 and NVIDIA GPU support, integrate with cloud logs/APM, and can be debugged by exporting disks and using GDB.
AI-written summary · about
chapters (9)
- 0:00Introduction and overview
- 4:58Unikernels vs Linux
- 7:02Demo of Go web server
- 13:07Real-world usage patterns
- 15:37Debugging and workflow
- 19:49Adoption challenges
- 22:41Mental model shift from containers
- 23:48Best use cases
- 24:52Auto-scaling without Kubernetes
related talks
38mUnikernels: The Next Evolution in Cloud Infrastructure
23mIt’s Time to Debloat the Cloud with Unikraft
33mUnikernels and Explorations
35mUnikraft: Unikernels Made Easy
38mUnik & Cloud Foundry Integration: Focus on IoT
32mUniK: Unikernel Backend to Cloud Foundry
39mUnikraft Has Arrived: The Lightweight, Fast, Usable Solution for Cloud Deployments
41mOSv: Probably the best OS for Cloud workloads you've ...