~vdupras/duskos-discuss

Dusk OS and networking

Details
Message ID
<34c58f06-03ca-48b4-9fa3-ae671d893631@app.fastmail.com>
DKIM signature
missing
Download raw message
Hello all,

In his new tutorial[1], Alex mentions that he believes that Dusk OS is an OS for
now rather than for the first stage of collapse. I have to say that I'm also
titillated by this idea and this makes me review my earlier stance with regards
to networking support in Dusk.

I've been looking at ethernet card datasheets lately and I also began reading
the TCP/IP guide by Kozierok (from No Starch Press). This all seems... doable.
Almost more doable than grokking NetBSD's code so that I know what I'd put on my
home router.

Because that's the whole underlying issue. I'd love to self-host in my basement
but I don't trust what I don't understand and exposing my home network to the
world is unnerving.

Wouldn't it be great if Dusk OS and Collapse OS websites were served by a PC
Engine Alix running Dusk? you bet it would :)

I know that what I've written above about NetBSD is completely crazy. It would
of course be less effort to learn enough about NetBSD to be able to trust it and
to use it in an effective manner than to write a whole TCP/IP+driver+http server
in Dusk. However, the former choice still implies a significant time investment
on a "technology branch" that I consider a dead end, that is the modern approach
to technology. The latter choice would be a lot more effort, but at least it
wouldn't be wasted.

And I have this other thought: what if Dusk's HAL[2] is inherently faster than C
code compiled by GCC? HAL code is a bit less convenient to write than C, but
it's still pretty okay, and Dusk makes it really easy to go back and forth
between Forth, C and HAL. So, I'm thinking, what if Dusk ends up having the
fastest HTTP server around? No C call convention overhead, no layer between user
code and driver code, you can hardly get more bare metal than that. What if the
idea caught on?

To be sure, I'm not about to start working on a new ethernet driver any time
soon because I want to keep my focus, but it's still ideas that circulate in my
mind.

But I think I'll try benchmarking HAL's speed compared to modern C some day
soon, just to see...

Regards,
Virgil

[1]: https://alexw.nyc/tech/duskos-1.html
[2]: and by that, I don't mean DuskCC, which I know compiled code that is much
less efficient code than "real" C compilers.
Reply to thread Export thread (mbox)