Hacker News Reader

Comment by pantalaimon | Hacker News Reader

How does NuttX compare to Zephyr?

NuttX is what you get if you ask “what would a POSIX runtime for small computers look like?”. Where most RTOS things (few of which are realtime at all, but whatever) emphasise hardware abstractions / device models etc., NuttX accepts that peripherals are hopelessly device-specific, and use of peripherals is usually hopelessly application-specific, and so doesn’t kill itself trying to have complete peripheral support for everything.

What it does have is a lot of the sort of infrastructure you need to build moderately to very complex applications, and an enormous supported device base. It also depends on essentially zero third-party code (there are a bunch of contributed / 3p feature libraries).

tl;dr - NuttX is kind of old-school, but it is lovingly maintained by some really nice people to quite a high standard.

https://nuttx.apache.org/docs/latest/introduction/about.html

I haven't used Zephyr much, but I think NuttX is a smaller, simpler OS than Zephyr? NuttX feels like a great way to learn about the internals of smartphone operating systems.

I would say it's a great way to learn about device drivers and hardware, because there's a lot less to learn than, say, the Linux DRM subsystem, and there aren't as many internal layers between your driver and the hardware. But modern smartphones run full-fat kernels with non-deterministic scheduling, not an embedded RTOS.

>But modern smartphones run full-fat kernels with non-deterministic scheduling, not an embedded RTOS.

I believe that is the beauty of it. Think something like the steamlink (streaming the video output of a steam game). Which turns into network transparent applications very quickly. Using a RTOS on the client side gives you amazing possibilities to lock the system down. Secure boot chains become a real possibility once you dont have to update that often due to the small attack vector. With NuttX having a simple framebuffer something like qubesOS but with remote VMs should be a real possibility. Especially since that is already happening with headunits in the automotive sector. Where one displayserver is handling both the safety critical tachometer as well as your multimedia stuff.

All of this not to mention, that the ARM processors in question dont have a secure coprocessor spying on you but Trustzone. This thread really gives me the warm and fuzzies. OP did something truly great here.

Reply on news.ycombinator.com