alt Hacker NewsCurious - how does one achieve this in Linux?
I assume initramfs-only with special purpose pid0 and only the modules needed statically compiled into the kernel?
What else would it take?
Curious - how does one achieve this in Linux?
I assume initramfs-only with special purpose pid0 and only the modules needed statically compiled into the kernel?
What else would it take?
The main slowdowns will likely come from device initialization and the bootloader.
Bootloaders need to initialize most of the devices and load the kernel image. Then they hand the control over to Linux which proceeds to re-init these devices again.
The userspace matters, but on recent computers it doesn't matter that much. You can get to sub-40ms with https://katacontainers.io/ That's a project that uses full VMs to run Docker images boot instead of kernel namespacing.