Hacker News Reader

Comment by norswap | Hacker News Reader

I'm a firm believer (ok, a wannabe believer) that one day the "whole goddamn corrupt temple" will collapse under the weight of its own complexity; and then we will go back to simpler, cleaner fundamentals. Maybe that will be the time of Minix. But maybe not, maybe it'll be the time of Mirage OS instead.

I really wonder what the overlap between Minix and Mirage are btw. Could Minix be used like Mirage as a minimal application-specific OS?

View on HN

The accumulated filth of all their drivers and patches will foam up about their waists and all the programmers and sysops will look up and shout "Save us!"... and we'lll look down and whisper "No."

I'm not an OS designer or OS programmer so I'm just arm-chairing here.

Mirage OS (also Haskell's HaLVm) are layers on-top of Xen. It would seem to me that the microkernel and exokernel designs wouldn't really be used the way Mirage is on-top of Xen except that Mirage / HaLVm could replace much of the userland on the microkernel or be its own os library on an exokernel.

The overlap I think you're seeing is that Xen acts a bit like a "microkernel" and Mirage is the userland implementing the lower-level interfaces provided by the kernel. Minix in this sense is a more complete picture but I don't see why Mirage or HaLVm couldn't be "userland libraries" for microkernels and exokernels.

I think it's a great idea actually. Build a small kernel with very few responsibilities so it's easier to prove safety, then construct the whole of userland on-top of a known type-safe language (Haskell or OCaml) and you get a potent platform!

I think it's a great idea actually. Build a small kernel with very few responsibilities so it's easier to prove safety

That already exists and was just open sourced! See seL4:

http://en.wikipedia.org/wiki/L4_microkernel_family#High_assu...

> I think it's a great idea actually. Build a small kernel with very few responsibilities so it's easier to prove safety, then construct the whole of userland on-top of a known type-safe language (Haskell or OCaml) and you get a potent platform!

While I like constructing a userland using a safe language over a small safe microkernel, I am annoyed by the hype this is getting because it's not new. Thinks like QNX, Symbian, GNU Hurd are based on this very concept. The only difference nowadays is the renewed interest in safer languages, which I suppose is only enabled thanks to Moore's Law giving us the headroom. It's also like everyone forgot the flamewars (Torvalds-Tanenbaum anyone?) about how microkernels were inherently slower. I'm sure that demon is going to reemerge soon enough, and we'll be back mucking around in C/C++ for performance. Just like 20 years ago. Those who don't learn from history...

Torvalds' argument wasn't that microkernels were slower though. It was that distributed algorithms are harder than shared memory algorithms.

Having not written a kernel, I can't judge those claims. It would be interesting to know if Minix has any "harder" algorithms than Linux because of this.

I think L4 already existed around that time, and got 10x or 100x improvement in IPC times, showing that microkernels could be fast enough.

Xen doesn't just act a bit like a microkernel, it is a microkernel.

You want lightsabers when everyone is using laser pistols. Good against remotes is one thing. Good against the living, that's something else.

While I appreciate the lyricism, I'm not sure this is an apt metaphor. After all, you can deflect laster pistol shots with a lightsaber, but you can't hardly deflect a lightsaber using a laser pistol.

But honestly, I'd rather learn to use the force. Better get that midichlorian shot.

Reply on news.ycombinator.com