When will MS replace the NT-kernel in windows?

[tom66]

The Windows NT kernel is actually rather decent, and certainly the last thing that should be changed in Windows IMO. The idea doesn't make any sense.

I'm not so sure about that.  The Windows kernel still has the unfortunate limitation where a SMB network drive locking up kills the entire filesystem driver until the SMB times out.  That shouldn't be possible.  There are a few single threaded locks in Windows that seem to be there more for simplicity than proper design.  It also doesn't handle high CPU counts as well as Linux, with multithreading being distinctly less efficient than Linux, and interprocess communication is a lot harder and less efficient too. 

In terms of raw stability, Windows and Linux are probably about matched, but if I could I'd definitely run a Linux-kernel Windows OS - provided it still ran normal Windows apps - which is probably not ever going to happen but we're discussing hypotheticals so...

[radar_macgyver]

@gnif thanks for the insights on that - my observations come from using an AMD GPU in a non-VM mode for home use (light gaming at most). The refreshing bit relative to NVIDIA was not requiring akmod for the display driver, which made kernel updates somewhat fraught since if the driver didn't build, I would have to spend a while rolling back to an older kernel and trying again. Haven't had to think about it since I switched to AMD.

I have seen my work computer (Win10, with an A2000 GPU) recover from a device driver crash - now I know how! Also, I do remember Level1 Techs covering Looking Glass a couple of years ago and wishing I could use it (work IT folks wouldn't let me). Amazing work!

[DimitriP]

Since Microsoft is now a services business, and don't rely heavily on Windows for revenue,

without "windows".... (other than  Dell, HP , lenovo and a few others getting out of the business of selling "computers"),   a lot fewer customers would have an "entry point" into the microsoft jungle "ecosystem".

"windows" is not what brings in the most revenue ...but it is what makes making money possible

 

 
 

[Nominal Animal]

As we all know, Apple replaced its original kernel(s) with XNU (for Mac OS X).  It has an interesting history, but the key point regarding the discussion at hand is that they still had to revamp the userspace-kernel interface completely.  For a number of reasons, I do not think that is possible for Microsoft to do something like that, at all.  Their historical baggage is too great, and investment in rejecting other approaches too high: it would be like trying to stop and redirect a juggernaut.  It would have to become a new, completely different company first.

Android and iOS have shown that new operating systems are commercially viable, but getting large existing software houses to port their software requires "subsidies" (incentives, support, et cetera); otherwise it will not happen.  On the server side, free/open source software like Linux and BSDs have already made it nearly impossible for traditional-proprietary-approach operating systems to compete –– all top 500 largest supercomputers and about 41.5% of all websites run Linux; about 84% of websites run an Unix-like OS –– so the ones left really get paid for maintenance, quality assurance, and support, instead of the binaries.  The fact that MS has these software houses already on board means any action that risks them leaving en masse could be a deathstroke for Microsoft the company; but also that keeping to status quo as much as possible, gives the company a big edge over any possible competitors.

Even though I don't want to touch Windows at all myself, I am happy that there are options and completely different development approaches.
Expecting Microsoft to replace their kernel with something completely different –– no matter what the perceived/assumed/believed technological or monetary or other benefits it would involve –– is unrealistic, I believe, based on the above.

[tooki]

As we all know, Apple replaced its original kernel(s) with XNU (for Mac OS X).  It has an interesting history, but the key point regarding the discussion at hand is that they still had to revamp the userspace-kernel interface completely.

Are you referring to replacing the Mach kernel in NeXTStep with XNU for Mac OS X (which is directly descended from NeXTStep), or replacing the classic Mac OS kernel with XNU? In the first case it was replacing one thing with a derivative of the same thing; in the latter, the kernel was replaced by replacing the entire operating system. I don’t think either of those is really comparable to replacing the NT kernel with the Linux kernel.

[tooki]

The last step missing is the replacement of the NT kernel with the Linux kernel.

When do you think this will happen?

Never, because doing so doesn’t make sense. It’d be a huge risk with no obvious reward.

[AndyBeez]

My opinion, other than the use case for domestic personal computers, who in the real power user world is using NT architecture? Even in the SOHO, NT is becoming a relic of a bygone era. A quick look around my personal tech zone here and... [pause]... there is my legacy laptop with Windows XP. I switch it on sometimes for retro gaming. Everything else, OSX or Linux. Businesses to campuses, personal devices to the cloud, the centre of gravity has moved parsecs away from the Windows black hole that crushed money in exchange for limited data security. The Linux genie is so out of the bottle that you cannot touch anything containing software that is not running a distro in one form or another. Controversially, Azure remains a cloud player only because it is heavily reliant on Linux under the hood(?). Replacing NT Kernel? Unlikely, as a MS will have gone broke long before a new "it's not Linux" kernel reaches the long term support phase. The OS world is also shifting away from the North American/IBM/MS model of cumbersome corporate software provision to an Asian model of faster, lighter, cheaper and adaptable. India or China is from where the next great OS will emerge.

Think NT Kernel, think steam locomotives. There are many enthusiasts who like to romanticise the past imperfect.

[madires]

The Windows operating system and NT kernel are quite fundamentally different to Linux which is based on a POSIX-like architecture.

One of the original goals for NT was POSIX compatibility. Another interesting thing is that NT is based on OS/2 (joint developement of MS and IBM) and VMS (MS hired a group of developers from DEC). Originally NT was also meant to be OS/2 compatible, but MS went for the Windows API after Windows 3.0 turned out to be quite successful.

[mendip_discovery]

Why would Microsoft dump the NT-Kernel?

That would alienate their customers. Do you know how much tooling depends on Windows?
You know, tooling that people use to actually do work. No I don't mean office. I mean CAD software and such.
There is decades of work in tools like Altium, Eplan, ArcGIS, Revit.

The makers of those tools should have started porting to Linux a long time ago. Linux is much more suitable to run CAD software compared to Windows anyway.

Linux has had an issue for years and it is much like the one that Apple has. It is not the OS but the user base. There are some rather toxic *nux users out there. I use it but I don't use it for my daily driver.

[DimitriP]

Think NT Kernel, think steam locomotives. There are many enthusiasts who like to romanticise the past imperfect

Lose the wheels. Antiques!

The OS world is also shifting away from the North American/IBM/MS model of cumbersome corporate software provision to an Asian model of faster, lighter, cheaper and adaptable. India or China is from where the next great OS will emerge.

 

[Nominal Animal]

As we all know, Apple replaced its original kernel(s) with XNU (for Mac OS X).  It has an interesting history, but the key point regarding the discussion at hand is that they still had to revamp the userspace-kernel interface completely.

Are you referring to replacing the Mach kernel in NeXTStep with XNU for Mac OS X (which is directly descended from NeXTStep), or replacing the classic Mac OS kernel with XNU?

The latter.  They did do quite large changes between Mac OS 7.5 and 8, too.

in the latter, the kernel was replaced by replacing the entire operating system. I don’t think either of those is really comparable to replacing the NT kernel with the Linux kernel.

You cannot simply replace the NT kernel with the Linux kernel.  If you could, then Wine would run perfectly.  Plus there is the large number of vendors providing binary-only Windows drivers, who would be impacted.

My point was that even Apple couldn't just switch to a better kernel; they had to revamp the userspace-kernel interface and APIs as well.

[nightfire]

Yes, Apple made a deep cut to get rid of lots of heritage by going OSX.

But seriously, from what we are already see in the cloud business of SaaS and other services, operating systems do not matter as much as some maybe 10 years ago. Nowadays the question is which API are available between pieces of software and how to configure them.

So with a timeline of say 2030 it should be quite possible for M$ to bring an operating system to market, like Windows ng (next gen), which basically serves as a launcher for cloud services and desktops and only allows installation of specially certified software. Professional Users then can rent the special Pro edition, where the basic desktop services are provided by cloud ressources (VDI/VDA). Installation of 3rd party software costs extra.
For manufacturers nothing will change in a big way, most of the good manufacturers are already providing some linux drivers for their stuff or the out-of-the-box drivers work.
Such a move also could bring driver development on linux to a better world,  because now some real commercial money is put into it.
 

[nctnico]

You cannot simply replace the NT kernel with the Linux kernel.  If you could, then Wine would run perfectly.

The kernel is not the problem for Wine. It is the dozens upon dozens of layers of software (userspace libraries) that sit between the application and the kernel. That is where the problem is for Wine; it is such a mess that it is next to impossible to emulate correctly. And I doubt Microsoft can re-create these libraries to a satisfactory level. Some backwards compatibility got broken in Windows 10 for example. Remember the guy from Sysinternals? He got hired by Microsoft because he understood more about the internals of Windows than the people working at Microsoft.

[tooki]

As we all know, Apple replaced its original kernel(s) with XNU (for Mac OS X).  It has an interesting history, but the key point regarding the discussion at hand is that they still had to revamp the userspace-kernel interface completely.

Are you referring to replacing the Mach kernel in NeXTStep with XNU for Mac OS X (which is directly descended from NeXTStep), or replacing the classic Mac OS kernel with XNU?

The latter.  They did do quite large changes between Mac OS 7.5 and 8, too.

Well they really just swapped out the entire OS. Other than the Carbon compatibility library (which required changes in application code, too) they couldn’t even run the same software. I don’t really think one can call this a kernel replacement, since it required replacing everything below it and above it too.

Not saying Apple’s achievement wasn’t important, but it’s nothing like the (nonsensical) “replace NT kernel with Linux kernel” posited by the OP.

in the latter, the kernel was replaced by replacing the entire operating system. I don’t think either of those is really comparable to replacing the NT kernel with the Linux kernel.

You cannot simply replace the NT kernel with the Linux kernel.  If you could, then Wine would run perfectly.  Plus there is the large number of vendors providing binary-only Windows drivers, who would be impacted.

I’m not saying one could; that was the OP’s idea. With that said, Microsoft likely could make it work; they’ve moved the Win32 APIs to multiple disparate kernels over the years.

My point was that even Apple couldn't just switch to a better kernel; they had to revamp the userspace-kernel interface and APIs as well.

”Revamp” makes it sound like they just did an update or refurbishment of those things, but what they really did was to dump the entire old OS and APIs overboard and replace them all with new ones. (Even Carbon was short-lived in practice, insofar as Carbon apps ran comparatively poorly and it was clear from the start that Cocoa, the native APIs from NeXTstep, were the future.)

[SiliconWizard]

Yes, it again wouldn't make any sense whatsoever, and if ever decided, would be a very bad move both for MS and for Linux. MS would get into this downwards spiral of never getting it right, and Linux would suffer enormously from this sudden too big influence of MS in its development, and then the lack of diversity, that others have also pointed out. Having options is good, even, and especially when some of the options look worse/better than others to *you*.

It wouldn't even be motivated by the need to target other architectures, as Windows is already there for ARM, and they probably have at least a side project to target RISC-V.

Then there's the licensing thing, I don't see how they could ever do with Linux's license.

I think some people also have an idealized view of Linux (and saying this while I have quit using Windows almost entirely) and an inaccurate view of what the Windows kernel is.

But anyway, I'm not sure what the OP's point really was. Maybe he also wants macOS to use a Linux kernel? The question was also "interesting", it was a "when", not even a "what if".
Back to point 1: having options is a good thing and drives improvements and innovation. Willing to corner everything to a single unified solution is never a good idea. And one reason Linux is great (while not without its deficiencies either) is that while being widely used, it's still independent. How can you ever imagine Linux (or any other project) staying independent while becoming a core part of their business? Knowing MS, it's, uh, hard to believe.

[nctnico]

How can you ever imagine Linux (or any other project) staying independent while becoming a core part of their business?

Linux is already a core part of every Android mobile phone out there. So basically, Linux is a core part of Google's business worth around 65 billion dollars per year. Then add the like of Samsung, Xiaomi, Huawei, etc

[Karel]

- MS maintains and offers their own Linux distro called "Mariner".

- MS has implented Linux into windows called "WSL2".

- MS has joined the Linux Foundation.

15 years back, nobody would have believed that but now it's a fact.
Today, nobody believes that, in the future, MS will replace the NT kernel with the Linux kernel.

To me, I can't look into the future but technically, I don't see any obstacles. Imho it's a matter of money.
I believe that, if it's cheaper for them, they'll do it.

p.s.: Two years ago MS hired Lennart Poettering (systemd inventor) who was working for Red Hat...

[PlainName]

p.s.: Two years ago MS hired Lennart Poettering (systemd inventor) who was working for Red Hat...

Windows is going to be even more fucked than it is now.

[Nominal Animal]

It is the dozens upon dozens of layers of software (userspace libraries) that sit between the application and the kernel. That is where the problem is for Wine; it is such a mess that it is next to impossible to emulate correctly. And I doubt Microsoft can re-create these libraries to a satisfactory level.

I agree; it is one of the technical reasons why a kernel swap would be extremely difficult.

Well [Apple] really just swapped out the entire OS.

I see your point.. perhaps it is better an example of how to switch OSes completely, and a simple kernel replacement unrealistic unless the kernels are already similar enough.

I’m not saying one could; that was the OP’s idea. With that said, Microsoft likely could make it work; they’ve moved the Win32 APIs to multiple disparate kernels over the years.

I don't think Microsoft could; the Linux kernel is just too different.  Perhaps, just maybe, to a microkernel; so they could add their own API and services on top of that.  I believe the internal changes needed to the Linux kernel to implement all Win32 APIs would just make it a different kernel altogether.

That said, I do not know the innards of the windows kernels; I have only inferred some of them, based on the APIs Microsoft has published.

[Nominal Animal]

Having options is good, even, and especially when some of the options look worse/better than others to *you*.

Exactly.

To me, this includes the wide variety of Linux distributions, which some feel is a downside and makes proprietary software distribution too difficult.  (I've argued about this earlier in different threads here, and while I can describe the techniques to overcome the differences, I still haven't created a practical example.)

(WSL and WSL2 I consider very similar to Wine: not everything will work correctly, and for some things you need to go to the real thing, for example via virtualization.)

I think some people also have an idealized view of Linux (and saying this while I have quit using Windows almost entirely)

True.  It seems to weird people out when I mention that very few Linux (kernel or other software) developers actually put any weight on the number of users they have, because it is irrelevant to them.  Being 'enticed' with "but you'd get loads of more users that way" reads like "but you get to shovel much more snow on your own that way".  It is an ecosystem, with rules to how it works (when it works; it does not always!), and it is nothing like one might expect or assume.

[CatalinaWOW]

There are two things going on here.  The technical problem and the social problem.

The essence of the social problem is the OPs demand for a switch to an environment that meets his needs and that he is familiar with.  His counterpart is easy to find in the Windows world.  I try not to be that person, but I'm not far from it.   I am not a software professional or a fanatic software hobby person.  I really don't care about kernels, or drivers, or such.  I care about simulations and models and photo editing and text editing and all of the end services provided by the machine and the software system.

For the most part the Windows system could handle all of my needs clear back at Windows 95, while at that time Linux wasn't up to many of those tasks.  Now Linux has largely caught up, but there are issues.  All large and powerful software packages have a steep learning curve, and while GIMP can do everything I need, I haven't gotten nearly as far up its learning curve as I have with the Windows counterparts.  MS commercial practices have driven me from MS Office to Libre Office, but I still can't use it as well as the old system.  And my investment in document automation is totally lost, and as nearly as I can tell in some ways not possible to recreate.  This list goes on and on. 

Both environments share a common problem, although the driver is different in the two environments.  Both want continual "improvements", most of which don't have much benefit to me as a standalone user.  MS needs to drive new sales, my purchase of an Office Suite 20 years ago is of no business benefit.  The Linux world spends an inordinate amount of time on new and shiny.  Everyone has their own almost compatible distro, with much emphasis on whether there are rounded corners on windows and other similar cruft.  I don't discount the technical improvements in both systems (larger memory models, automatic HW configuration, etc.) but to be honest these are luxuries, not core requirements for my use case.  I can also see the need for better security features, but I also can solve that problem pretty easily by isolating my system. 

A long winded diatribe, but both environments would benefit from a step back and figuring out why anyone wants their product at all.  Apple and various hardware vendors have decided that the large market is web shopping and social media.  I don't think they are wrong, but neither of those really requires much of an OS.

[JohanH]

p.s.: Two years ago MS hired Lennart Poettering (systemd inventor) who was working for Red Hat...

Windows is going to be even more fucked than it is now.

I don't understand the hatred for this guy. And I've followed the whole systemd debacle since the start, read his blogs etc. I guess, he was too clever and too outspoken for the mob. It for sure brought out all the ugliness you can imagine in the circles; the amount of trolling and hatred by his critics I haven't seen since (except in politics). But what it takes to have development and invention (and society) make huge jumps is guys like him that are visionaries and able to realize the ideas, too. I for my part admire him. If anything, he will do good things at Microsoft.

[Karel]

p.s.: Two years ago MS hired Lennart Poettering (systemd inventor) who was working for Red Hat...

Windows is going to be even more fucked than it is now.

I don't understand the hatred for this guy. And I've followed the whole systemd debacle since the start, read his blogs etc. I guess, he was too clever and too outspoken for the mob. It for sure brought out all the ugliness you can imagine in the circles; the amount of trolling and hatred by his critics I haven't seen since (except in politics). But what it takes to have development and invention (and society) make huge jumps is guys like him that are visionaries and able to realize the ideas, too. I for my part admire him. If anything, he will do good things at Microsoft.

I guess if he should have respected the general "Unix" philosophy that tools should do only one thing and do it good,
the reception should have been much better. Instead, he created a monster that wants to do and control everything.
What will be his next step? Creating a Linux registry for all settings like windows does?

[JohanH]

I guess if he should have respected the general "Unix" philosophy that tools should do only one thing and do it good,
the reception should have been much better. Instead, he created a monster that wants to do and control everything.
What will be his next step? Creating a Linux registry for all settings like windows does?

That's just repeating the same conspiracy crap. Can't we just go forward now.

[PlainName]

I guess if he should have respected the general "Unix" philosophy that tools should do only one thing and do it good,
the reception should have been much better. Instead, he created a monster that wants to do and control everything.
What will be his next step? Creating a Linux registry for all settings like windows does?

That's just repeating the same conspiracy crap. Can't we just go forward now.

Isn't it looking at what is actually hitting the metal?