Realtime != high throughput. It just means deterministic throughput. FSVO deterministic.
Show me people running big farms of servers running these operating systems where even single-percentage computational overheads really matter.
(added:) The reason for this is that it costs one hell of a lot flipping your page tables and flushing your TLBs every time you have to switch to ("pass a message", whatever) to a different subservice of your kernel.
(also added:) Oh and interestingly many (most?) users of OKL4 go on to host Linux inside it because, hey, it turns out that doing all your work in a microkernel ain't always all that great. So 90% of the "kernel" work in these systems is happening in a monolothic kernel.
Other contenders include eMCOS and FFMK, though those are obscure.
That said, I don't even understand the logic. HPC clusters where single-percentage overheads really matter are an extremely specialized use case, so of course COTS u-kernels might not cut it. Where's the shocker here?
Response to added: Not necessarily with message passing properly integrated with the CPU scheduler.
Response to added #2: Hosting a single-server is a valid microkernel use case. What's your problem? Isolation and separation kernels are a major research and usage interest.
I'm not even really talking about HPC, just the massive datacentres that run everyone's lives. All for the most part running monolithic kernels. I doubt the thousands of engineers who work on such systems consider the "huge monolithic kernel" "undebuggable". And I don't see examples of microkernel OSs that are able to cut it in these circumstances.
Even in a mobile device, you don't really want to waste battery doing context switches inside the kernel.
Microkernels have their place, but believing that the world that chooses not to use them are just clearly dumbasses is bullshit dogma.
You appear to be assuming (not being a mind-reader, whether you actually are or not is of course unknown to me) that QNX would automatically be used in server farms if it was high throughput; and, since it's not visibly used there, it is not high-throughput.
(As an aside, I'll grant that even a high-throughput microkernel seems likely, to me, to have a lower throughput relative to a more tightly-coupled monolithic kernel. That's just one of the architectural trade-offs involved here.)
As I see it, there are technical (e.g. hardware drivers, precompiled proprietary binaries) and social (e.g. relative lack of QNX expertise = $$, proprietary licensing) reasons for many people to choose one of the more popular OSes, running monolithic kernels.
I can't say what's technically superior, but even if QNX was, nobody's a dumbass for choosing something else -- and I don't think the fellow you're replying to was saying so. There are, of course, reasons and trade-offs.
An OS's adoption is a social thing, and proves nothing technical about it. If it wasn't for licensing (a social problem), BSD might have taken off, and Linux been comparatively marginalized.
Entertain for a moment the idea that someone's rationale in choosing to deploy a given OS lies deeper than the 1-dimensional rubric you're suggesting, and instead may have something to do with questions like "how easy is it going to be to support this?" and other network effects.
You're getting all red in the face using some really dubious arguments to back you up here.
(response to 2:) Er, so bypassing the microkernel for the vast majority of your work is a vindication of the "microkernels are just better" line is it?
High throughput, mister, high throughput.
Realtime != high throughput. It just means deterministic throughput. FSVO deterministic.
Show me people running big farms of servers running these operating systems where even single-percentage computational overheads really matter.
(added:) The reason for this is that it costs one hell of a lot flipping your page tables and flushing your TLBs every time you have to switch to ("pass a message", whatever) to a different subservice of your kernel.
(also added:) Oh and interestingly many (most?) users of OKL4 go on to host Linux inside it because, hey, it turns out that doing all your work in a microkernel ain't always all that great. So 90% of the "kernel" work in these systems is happening in a monolothic kernel.