Show HN: Minoca OS – A new lightweight kernel, completely from scratch | Hacker News Reader

Show HN: Minoca OS – A new lightweight kernel, completely from scratch

minocacorp.com

168 pointsEvanGr42 comments

I suspect novel proprietary kernels are a very difficult business niche. I'm unaware of any new commercially viable OS kernel that has come along in the past decade...maybe even longer. Linux (and to a lesser degree, BSD) seem to have a very firm lock on the market, from quite small devices on up to very large ones. Certainly, there's still room at the very bottom (devices smaller than RPi) and at the very top (what little big iron remains) that isn't owned completely by free and Open Source systems...but, it's looking like a pretty tight squeeze, to me, for a system that targets the small end of Linux' market, and does so without some dramatic differentiation (the mentioned 5MB size is entirely do-able with Linux, even modern Linux...old Linux kernel versions could fit on a floppy with room to spare for utilities).

It's not that I don't like seeing new kernels and new ideas. I'm just not seeing a path to profitability for something like this (but I thought Dropbox was a bad idea, so I may be stupid).

So, what's the market niche for this that isn't better served by an Open Source platform?

Thanks for the feedback. We're optimistic about the idea of focusing on what can be gained by starting from scratch. Specifically by creating a modern API between the core kernel and device drivers, we think we can make interesting advances in areas like system-wide power management, component level servicing/updates, and overall "debuggability". We believe these areas will become even more important as general purpose operating systems are integrated into more embedded devices.

The problem is that as long as the kernel is proprietary, people have no reason to trust it to run that kind of hardware.

There may be another exception: very specialized OSes in critical environments like planes or satellites.

These environments generally need a real time OS with a small footprint (for auditability, or even prove a part of the code).

Yeah, but you need to design an OS from scratch to supported these hard realtime and safety requirements. There are a few ones out there (VxWorks, Integrity, ...). But the linked OS seems more to focus on the general purpose OS role.

QNX is a commercially successful micro-kernel being used in automotive industries.

QNX is 30+ years old, and is effectively defunct except in a few niche markets after being acquired by Blackberry (and was losing market share to Linux for decades before that). QNX is not an example of a new and successful proprietary kernel competitor to Linux, despite being well-regarded and pretty neat technology.

It was. And it's heavily on the decline, since it's substituted for Linux in a lot of new projects. This underlines SwellJoes assumptions that new operating systems have it very hard.

And is about 35 years old... GP mentioned 'last decade'

It looks like this is a closed binary blob. Disappointing, since I'm curious to see what exactly it does differently.

I'm also not entirely sure why I would want to use this over Linux, from reading the website.

Better debugging, commercial support, stable/better driver API I would assume. They explicitly state they take away the pain of constantly maintaining patches.

http://www.windriver.com/products/linux/

That page is really missing a description who is maintaining it (who is the team, how big is it, where is it located?), especially if it is marketed as a commercial offer. I don't think anybody will download binaries from some unknown individual and build a commercial product on top of it. In Germany you could even get sued for a page like this.

It would be really awesome if you could provide a quick and easy way to test it. If this is really so small one possibility would be to port to run on one of the many JavaScript machine emulators like:

http://bellard.org/jslinux/

It would let you try the OS in the browser.

The "Product" page gives a much better explanation of why we should care then the linked "Download" page.

Interesting. How much "smaller" is this in practice than a stripped down Linux image?

It boots to a shell in about 5MB of RAM, with networking and USB. Both Linux and Minoca can get smaller than that depending on how much you're willing to strip. The goal was to start lean and clean and then go up or down depending on specific needs. (Disclaimer: I work on Minoca OS).

People used to use eCOS and RTEMS for this stuff. Plenty of components available for each, too. Anyone wanting minimal OS should check them out.

Can it provide any real-time guarantees? The main reason why I would change OS on the Raspberry Pi 3 is to get real-time behavior.

Not at the moment, but our interrupt overhead and number of background tasks is so low it's something we're thinking about. Is there a specific scenario you're trying to enable?

I don't mind the "heaviness" of Linux on the RPi but I cannot rely on its user-space interrupts for GPIO use from userland. Instead, I have to write kernel software or give up on guarantees. So far it hasn't been much of a problem, but if I wanted to do more tightly timed things, I'd need to use a real-time OS.

Meanwhile, it's nice having the support of lots of Linux software so it's hard to justify switching to a non-real-time OS with much less support.

I've noticed, browsing the RPi forums, that people are disappointed that they can't get real-time behavior from Raspbian.

I don't know about the grandparent, but I'd imagine anyone interested in using something QNX (http://www.qnx.com/products/neutrino-rtos/neutrino-rtos.html) would be interested in Minoca, if it became a real alternative for a RTOS.

It'd be interesting to see more information about licensing for Minoca, as I could see it as an interesting alternative for small projects on the RaspberryPI.

Cool! Can you describe a user who would see this as a perfect fit?

We're systems people, so we can help folks building new hardware devices with tasks like bring-up and sustained support. There's a stable kernel API, which makes component level updates within the kernel possible.

It looks neat, but is this something I'd use a RPi as an overpowered arduino for? What does this offer?

Yes, you could use Minoca to manage GPIO and other low level hardware. One area where we're working to excel is power management. There's so little in the OS that we're able to go very idle, and stay there for long periods. Another area we're focused on is component level servicing: being able to update components within the kernel while maintaining a high degree of stability.

To user mode we'd like to present a largely compatible interface, while in the kernel we want to innovate using new designs and a clean slate.

Is this a commercial product? If so, what's your business model? This looks really neat by the way.

Just a quick review of the site - the "Product" page - shows a lite version and a pro. Lite has debugging symbols built in that can't be stripped. It is compiled at a lower optimization level than pro.

It says for Access to Source: Lite - Limited, Pro - Unlimited

Not sure what that means though, I didn't dig very far.

Found the above here: http://www.minocacorp.com/product/ - scroll down a bit to find the product comparison. I quickly checked each top level page and did not see any prices. Unless I missed something it looks like it's a contact us and we'll tell you situation wrt pricing...

(emphasis mine):

> It says for Access to Source: Lite - Limited, Pro - Unlimited

In the software world I'd consider anything less than BSD/MIT licensed source code limited.

> In the software world I'd consider anything less than BSD/MIT licensed source code limited.

Or GPL. Any free license would be acceptable.

GPL is by nature limited. All viral licenses are.

All software licenses are limited. Im limited to following the BSD license conditions in BSD licensed software or I won't be able to redistribute the software.

The only non-limited license is the non-license that goes under the name public domain, but then I am again limited to only use software that was developed 100 years after the developers death. Some punched cards "programs" from IBM in the late 1800s should thus now be the first unlimited "software" which everyone is free to use.

You're not limited as a user. As a developer, the only "limitation" is that you have to provide the code with the binaries.

Ooh I didn't see that menu item because the link was to the download tab and product was left of it

Yeah, the site seems to get right to the point on how to download and install it so I can see why the OP started with that link. As mentioned above price and details on what actually is different between the two versions is a bit lite on details but seeing the OP's comment below mine suggests they are working on it...

Looks like an interesting project though. Going for Posix compatibility and having a working sh, gcc and python to start with can only help garner interest.

We're hoping it can be, but we're also considering open sourcing it.

You could always put it under the AGPL 3 (or later). Your prospective customers would probably still want to buy a less restrictive license from you, because they are irrationally afraid of the GPL, especially GPL3, and even more especially the AGPL 3. But free software would still benefit from the source code being free.

You can open source it, and it still be a commercial product.

What's the hardware compatibility like? If you're open to it, there is a market for something that will boot with the bare minimum non-specialized generic drivers on all hardware into a shell or some form of UI.

We currently run on x86 PCs and a number of ARM SoCs (the linked page should give you an idea of which ones). Adding support for new SoCs and boards is pretty straighforward. Feel free to reach out to me directly with what you have in mind: evan AT minocacorp.

Very cool! Do you think this has an application outside of embedded, like microkernels for lightweight containers?

This project looks promising, but I can't find the link on the website to download the sources.

Reply on news.ycombinator.com