Gartner: Windows collapsing under its own weight; Radical change needed

Windows will need different kernels for different applications. Yeah that makes sense. Good luck with that Redmond.

Gee, where's the "Linux kernel is so lean and mean that it's working in every application people can throw at it!" article?

Bob -

You know that's not actually true, right?

Compared to Windows, yes, but compared to any reasonable OS, no. What we have is a modular kernel that is able to be many different things according to our needs. Very powerful and very flexible.

@Bob:

AFAICT, that's the main point of the Completely Fair Scheduler.

Interesting how Windows is splitting the OS kernel, while Linux is consolidating.

As far as kernels go, the following is an interesting, unorthodox opinion piece:

http://ols.108.redhat.com/2007/Reprints/lameter-Reprint.pdf

The old "we gotta have a microkernel" thing refuted pretty solidly. Heck, even Apple took NeXT's Mach 2.0 "microkernel", and that was a pretty fat "micro" kernel.

dino, you're technically correct, but this is the expert tech press we're dealing with here. So that makes Bob's approach the correct one.

Funnily enough, when you read the wish lists in the article, it's clear they are describing Linux, and perhaps FreeBSD, which has also been customized for a wide range of platforms and applications.

To run on a cell phone, Windows has to have an entirely re-written kernel. Linux uses the same kernel, just compiled for that environment without any fluff.

To run on a mainframe, ..., oh, right, Windows doesn't do that. Linux uses the same kernel, just compiled for that environment.

To run on a DEC Alpha 64bit computer, Windows 95 had a version if you can find it. Linux uses the same kernel, just compiled for that environment.

To run on a SPARC, ditto. Wristwatch, ditto. Zaurus, ditto.

Anyway, that is what I was trying to point out.

Scheduling is another matter, as Dino has pointed out many times in the past, configurability is increasing in Linux as experience is gained. This will only increase the effectiveness of the "one mainline Linux kernel" in comparison to the "write many different kernels" of Windows.

Quoting: "write many different kernels" of Windows.

Giggle. That should be a roaring success. Look how well the richest software company on the planet has done at managing a single kernel.

TC and Bob --

I think having many different opportunities will open new Vistas for Microsoft.

> will open new VISTAs for Microsoft.

One can only hope.

:D :D :D

> To run on a DEC Alpha 64bit computer, Windows 95 had a version if you can find it.

Actually that was nt - it was a 32 bit OS on 64-bit hardware and it was quite unimpressive.

When I was at UC we had taken delivery of some Dec Alpha boxes with the much hyped nt installed, played around with them for a bit, and ran some benchmarks.

After the novelty wore off and it became boring, the boxes were scrubbed, and digital unix installed. The same benchmarks were 3X faster on each unix box, compared to when the same box was configured as a windows pc.

I think we're getting our terminology confused with Gartner-speak. In Gartner's world a kernel is a binary blob you get from Microsoft and you run it and hope nothing breaks because if it does there's no way for you to fix it or debug it sans source code.

In our world we have this thing called the kernel source code and we can play with the configuration file and compile our own binary blob. And people can create their own kernels for cell phones or mainframes or routers or megaclusters using the same source code.

That said, I have my doubts about Microsoft producing and supporting multiple kernels given their record with one kernel - one platform.