Cut your Cloud Computing Costs by Half with Unikraft
summary
Unikraft claims concrete wins: cut cloud costs by half on an EC2 experiment by running Unikraft unikernels that match a large-instance Linux nginx on a medium instance. Images are a few megabytes, memory is 2–7 MB, and nginx throughput reaches ~300k req/s. Builds use an auto-porting flow (binary and source modes) with ~146 syscall support, boots can be a few milliseconds (microVM ~9 ms, solo5/firecracker single-digit ms), and a custom SHFS serves files in ~300 cycles.
AI-written summary · about
chapters (23)
- 0:00Intro
- 0:14The Private/Public Cloud
- 1:09The Problem Part 2: Time
- 2:46Unikernel Power
- 3:41Unikernels in One Slide
- 4:50Design Principles
- 5:46Doing it with existing unikernels?
- 6:24Specialized API Example
- 9:06POSIX Compatibility - Two Approache
- 11:08Syscall Support Top 30 Debian Popcon Apps
- 11:59The Intuition
- 12:28Syscalls: What's really needed
- 13:24What Syscalls do I Really Need?
- 14:43SQLite: Manual vs. Auto Port
- 15:31Image Sizes vs. Other Projects
- 16:10Unikraft Boot Times
- 17:06Minimum Memory Requirements
- 17:29nginx Throughput
- 18:05Redis Performance
- 18:24Boot Times - Different Allocators
- 19:00Redis Throughput Different Allocators
- 19:49Filesystem Specialization - SHFS
- 21:48Towards Seamless Integration & Deployment
related talks
30mBeyond Orchestration: The Cloud Native Runtimes Ecosystem for Performance and Security
10mLightning Talk: Millisecond Scale-to-Zero and the Quest to Never Pay for Idle Again- 10mLightning Talk: Millisecond Scale-to-Zero and the Quest to Never Pay for Idle Again
44mUnikernels and Hyper Elastic Clouds
34mLook ma, no OS! Unikernels and their applications
38mUnikernels: The Next Evolution in Cloud Infrastructure
24mUsing Unikernels to deploy real-world apps
25mExtremely Fast and Efficient NFV with Unikraft