Glass Elevator, Idea #3: new Kernel

Haiku kernel (R2) based on the so-called pragmatic universal model based partly on fuzzy neural networks controlled by a genetic algorithm.

Neural installation kernel in the processor, the rest are traditional computing units with additional instructions

The concept of neural transformations = self-maintenance, auto-updates, self-determination of AI

Please post the message!

1 Like

I am probably too old for fancy things like this. I better like programs which i can understand and which doesn’t try to be smarter than me. That causes always a disaster.


Skynet likes this.


This post reads like it was generated by an AI.


this is just a theoretical concept, the first neural network processors are just appearing.
I faud a some hindu people “AKAdemiks” work on that, btw no one know how to programing that.

Please start your own crazy project, we have enough work just making a normal OS already :slight_smile:

Also, saying “no one knows how to program that” is not quite right, there are neural net processors in today’s CPU hardware already (I have one in my current laptop). It shows up as a PCI device and Intel provides a toolkit for it

But this is not to replace the main CPU or run your kernel on. It is some accelerator device, just like there are for 3D rendering, for video encoding/decoding, and so on. So if you have an algorithm where you need a neural network, you can do it reasonably easily on a CPU or a GPU, but now you also have the option of this dedicated hardware, which will do it a bit more efficiently.


Please start your own crazy project, we have enough work just making a normal OS already

Challege Aproved, now a have some time for this crazy idea :wink:

Ewww kernels are so 80s. Lets remove it already and move on…

Taking the opportunity to share my own idea for a “grand project” after R1, I would replace Haiku’s kernel and services with a new core system using the Genode OS framework. This mirrors how GNU intends to replace the linux kernel in due course with the HURD. The advantage is that Haiku switches to a next generation “capability based security” OS. In fact this is not an imaginative idea at all: Android is going this way also, switching out the Linux kernel in favour of Fuchsia.

I see this as a useful path to making Haiku more of an EU desktop OS. Genode represents teutonic engineering, but their own Sculpt desktop looks rather grey and drab. Be was founded by a Frenchman and (with Haiku) represents Gallic style and flair. Imagine combining Genode’s guts with Haiku’s desktop for the best of both worlds.

1 Like

Haiku already is a EU OS pretty much :slight_smile:

1 Like

I would bet on LLVM and other new utilities replacing GNU, before Hurd gets anywhere close to replacing the Linux kernel.


I suggest replacing the Haiku kernel with the Windows NT kernel so that we can directly use all proprietary drivers from hardware vendors without all the hard porting work.
Going with the most widely used kernel would give us a safe future with new features coming from Microsoft every few months that we can simply use in Haiku.
Additionally,we could make Haiku usable on more architectures easily,at least support for ARM64 is already integrated in the Windows NT kernel.
Maybe that step could even enable running Windows applications without a compatibility layer like Wine,and more reliable.

Sorry for trolling,but I had to do that now,with all those alternative kernel ideas coming up every now and then.
Seriously,we have our own great kernel which is the result of more than 20 years of work from many people,and we shouldn’t throw that away.
Haiku runs fast,even on older hardware,and while we still don’t have accelerated graphics,the software-rendered graphics often works better by default than on Linux or the BSDs.
Our network drivers are far better than FreeBSD,let alone Illumos or other niche systems,only OpenBSD can compete here,but there it’s too difficult to setup.
Instead of starting again from zero,I think we should continue from where we are and focus on improving hardware support even further and maybe copying some useful features from other kernels without replacing it completely.


But… they are FreeBSD drivers :’(


Aren’t they mostly OpenBSD drivers now?
Both Haiku and OpenBSD support Wifi a/c while FreeBSD doesn’t.
Some network chips are not detected or detected but can’t connect on FreeBSD,while they work perfectly fine on OpenBSD and Haiku.

1 Like

Haiku uses drivers from both OpenBSD and FreeBSD.

We reuse both their drivers, our ethernet drivers are also mostly FreeBSD. But it’s kinda wiers to say “our drivers are better” without also specifying “which we took from OpenBSD”

Haikus network stack cant compete with either if those OSs, and it doesn’t need to either

1 Like

There are no good quality open source and free software-compatible license NT kernel implementations yet. ReactOS kernel is terribly unstable and full of bugs. Using proprietary drivers for Windows also have a problem that most of them are distributed by Microsoft with Windows Update and not directly downloadable. It may cause legal issues by extracting driver binaries from Windows distribution or Windows Update data.

1 Like

As things stand I see Glass Elevator as merely a talking shop at this stage: a “safe space” to discuss ideas that may crystallise into a future roadmap once R1 happens. Even then it might be more of a parallel fork than an immediate successor: as we presently see with our legacy 32 BeOS compatible and modern 64 bit versions.

I agree that the practical work done thus far is a first class achievement as many reviewers of the latest Beta attest. But I disagree that our resect for this work is reason not to talk what may come after. As implied by the amount of time to get this far, even when we agree on this nebulous future Haiku we refer to as “Glass Elevator”, it may take 20 more years to bring it about.

A fork that shares 100% of the code in the same repository and is developped and maintained by the exact same team? Now that’s a strange idea :slight_smile:

1 Like

From the point of view of a humble user like myself - one who is pretty unlikely to build Haiku from source - it feels like two branches are on offer to download even if they are technically just differing builds of the same thing. What they offer the prospective user is different: 32 bit might as well be “openBeOS, for binary compatibility”, 64 bit is “Haiku, inspired by BeOS”.

We can probably leverage the existence of two builds as a way of selling the OS: the 32 bit version takes the brickbats of merely being a recreation of a quarter century old OS; allowing us to present the 64 bit version as a modern “legacy free” version: all at no extra effort. That’s marketing for you!