I notice that you don't balance how many times scientists have been wrong against how many times they've been right. What do you suppose a scientist's wrong:right ratio is as compared to a non-scientist's?

That is such an idiotic statement that I won't even bother continuing the discussion. This link is the wikipedia page.

Perhaps you should actually read the link you specified. Linus himself describes the term "hybrid" kernel as simple marketing, which is exactly what I said.

Pulseaudio is a brain damaged [beastwithin.org] piece of software and one of the first things to be removed in any distribution.

And yet, it's not removed seeing as it's still around and still quite popular.

Yet for machines like home computers it is simply not possible. It is only with the relative advance of hardware that the microkernel can actually get close to a monolithic kernel in terms of performance.

This is not true. The L4 kernel ran Linux as a guest OS with 5% overhead. This was the birth of hypervisors like Xen, which are really just a sort of microkernel. Virtualization is everywhere now, and no one seems to be complaining about overhead. If you can run a virtualized host with so little overhead, native execution is at least as fast as your guest.

It's perhaps too easy to introduce performance problems via a poor choice of decomposition, but that doesn't entail that decomposed systems must necessarily perform poorly.

It is this performance consequence that made Windows NT, originally designed on a microkernel architecture, move towards a hybrid kernel.

Poorly designed systems perform poorly. The NT kernel might have been a nice kernel design from some people's standpoint, but then people thought the Mach kernel was a good microkernel too. Both opinions are incorrect.

There are far more Mac and Windows installations than all Linux distributions combined. These are all microkernel or hybrid-kernel architectures.

There is no such thing as a hybrid. You are either on fire, or you are not. You are either a microkernel, or you are a monolithic kernel. Mac's may use the Mach microkernel, but it's a poor kernel and the BSD subsystem really consists of most of the system calls, which all execute in kernel space. This is a monolithic kernel that has a poorly designed microkernel as its ancestor.

Systemd is all about marketing, and nothing about engineering. It too will fail and be replaced, just like PulseAudio by ALSA.

Except PulseAudio hasn't been replaced, it's still used by most distributions.

In the end the real debate was HOW to accomplish the modularity, not whether to make the kernel modular or not

Right, so make the kernel modular via isolation which provides fault tolerance in your most critical piece of code, or make it monolithic, ie. everything lives in kernel address space.

There is no reason on earth that an init system would need a specific journal daemon

There is no reason on earth device drivers need to live in kernel space either. Performance arguments are simply false, and this point has been disproven many times over.

Of course arguments and hard data aren't meaningful in these discussions, and monolithic has clearly won in terms of marketshare. Once again, why fight the tid of history instead of being more constructive? You are going to lose.

But systemd isn't actually monolithic, it's monolithic by fiat because the daemons refuse to play well with others, which (again) is against the Unix Way.

Microkernels are arguably more "Unixy" than monolithic kernels. Each device driver is simply a process that has a well-defined stream interface that can be piped to any other process, not just the kernel itself. Microkernels are Unix taken to the extreme.

So again, this argument failed for microkernels, so why should it succeed here? Perhaps some core functionality, not just system calls, should also be in some monolithic service and not a set of composable subsystems.

It's taken until recently for Minix to become even vaguely usable as anything other than a learning operating system, it's lagged behind everything else in terms of features always.

Tanenbaum and others weren't arguing that Minix was the microkernel OS that should be selected, merely that some microkernel should be preferred over monolithic kernels. The high performance L3 and EROS microkernels both existed at the time, albeit in early stages like Linux.

Yes, that is an excellent reason to add even more vulnerability vectors!

Granted, but more granular fault isolation wasn't convincing when Tanenbaum and Linus were arguing microkernels vs. monolithic kernels, so why should it be convincing now? I'm certain your other complaints are fixable given the current framework, assuming there aren't other mitigating issues.

I have no opinion on systemd. However, more granular fault isolation wasn't convincing when Tanenbaum and Linus were arguing microkernels vs. monolithic kernels, so why should it be convincing now?

Every condemnation leveled against the monolithic systemd are just rehashed arguments of monolithic vs microkernels. Monolithic kernels clearly dominate, and chances are systemd will similarly dominate, so instead of wasting your time battling the tide of history, perhaps you should be more constructive.

What a lot of people are concerned about is that this entirely new and largely untested (in the 'wild', as it were) and very very large, complex piece of software which runs at a very very privileged level in the operating system is going to become the main source of security vulnerabilities in Linux.

Linux has almost two orders of magnitude more code than systemd, and it changes all the time. Security vulnerabilities are far more likely to be in the monolithic kernel.

It's still limited by the FPGA's gate count, which is pretty low by CPU standards.

We cannot build a computer that can model a bug's brain activity

Actually, I believe IBM emulated a rabbit sometime in the past couple of years.

You can have yearly releases as long as you're willing to ruthlessly cut features that aren't sufficiently stable. If frequent updates are more important than features, then that's achievable.

The problem would be if marketing had a hand in both direction AND quality control. That's the recipe for disaster.

The demo site uses frames. FRAMES.

This is the researcher's website. A RESEARCHER. Who cares if he sucks at web design? Ur/Web can generate any HTML you want.

That gave me some hope for the world. At least some stood up and said "No" and likely ended their careers over it.

Yup, it's a grand, sickening Milgram experiment writ large.

I think that it's more protective of citizens to behave in a way that isn't morally reprehensible.

That's an excellent and underappreciated point. The only difference between a morally reprehensible government-sanctioned action against a terrorist, and an action against you is an easily manufactured excuse. If morally reprehensible actions were never permitted, then the citizens need never fear their own government, which was the whole point of the constitution to begin with.

you can separate them and ask them questions then torture them when their answers don't match.

Except that doesn't work, because people being tortured will say anything to make it stop. At no point when they change their stories can you be certain they're now telling the truth. Even if their stories suddenly match, it could be a complete fluke, or as a result of the interrogator asking leading questions. Torture is useless.