LWN: Comments on "The future of the realtime patch set"

The future of the realtime patch set

kramlie — Sat, 28 Feb 2015 10:54:04 +0000

So LF and OSADL are looking for companies to fund the project, but what about donations from users? Of course they can't individually provide the money that a company can, but put together it might add up. The RT patch set has many users, for instance in the audio production scene.

Personally I'd very much like to contribute, and a simple donation button would make it easier.

The future of the realtime patch set

WMW — Fri, 19 Dec 2014 07:34:52 +0000

Many Thanks to the people who have done a very good job with PREEMPT_RT so far.

A reason why nobody wants to fund the further development may be, that the realtime patches are deeply integrated into the linux kernel and no one has control over linux kernels further directions. From the point of view of machinery builders the claim "Linux for the desktop" sounds like poison for real-time applications. Otherwise machinery builders have to deal with functional saftey and therfore with validation and verification of the OS used. Very difficult with 17 million LOC of kernel. They appreciate small rt kernels.

Nowadays there are better solutions out. Multicore SoCs are very cheap. So it seems better to partition the multicore for mission critical real-time, safety and linux applications. You can use unmodified mainline linux if you choose a core with hypervisor. Particulary if you have a product life cycle of 10 to 25 year, which is very common in machinery industry, this promises a clearly better maintainability.

I'm sorry to say this, the rt patch set is a really nice und sophisticated piece of software but has limited value for machine and machinery builders. It may be well suited for gadgets, but who needs real-time there?

There is definetely no market and therefore no funds for real-time patch set.

Solaris realtime OS?

liam — Tue, 04 Nov 2014 00:50:46 +0000

I'm not sure you're being completely genuous with that reponse.
RT certainly doesn't necessarily mean high performance (rarely, in fact, but not impossible in certain circumstances as, iirc, thomas gliexner has previously mentioned that he expected the preempt-rt network stack to provide even more throughput than mainline---I don't think this ever came to pass), or very short interrupt handling, but it does mean that you can offer guarantees of service, which mainline linux can't offer 100% of the time (always ignoring hardware failure).
Unfortunately, solaris has long had the advantage of extensive, and cheap, tracing, and have used this to be able to track down sources of excessively long preempt-disable.
LITMUSrt has a similar facility, but not the resources, or the purview, to do what solaris was able.
Your last line isn't applicable to the topic at hand, though it is certainly true in so far as it must be true.

Solaris realtime OS?

dlang — Tue, 28 Oct 2014 00:58:37 +0000

Anything can be "real time" if the deadlines are long enough.

And if the system isn't fast enough to get the work done, it's got no chance to be "real time"

Solaris realtime OS?

liam — Tue, 28 Oct 2014 00:33:38 +0000

Who said anything about faster?

Solaris realtime OS?

vonbrand — Mon, 27 Oct 2014 23:57:08 +0000

In 2000 Linux was so much faster on the same machine (SPARC) than Solaris du jour it was not funny anymore. Real time it wasn't, not by a very long shot.

The future of the realtime patch set

liam — Mon, 27 Oct 2014 08:29:51 +0000

Always great to see projects that improve the quality of the Linux kernel getting their due.
So, it looks like one area where Linux will never catch up to Solaris will be in its ability to provide latency distributions. Yup, apparently Solaris has had real-time capabilities since 1993.

The future of the realtime patch set

speedster1 — Mon, 27 Oct 2014 07:16:36 +0000

Disappointing to see this... why doesn't the Linux Foundation step in here? This seems like a good investment for those who actually have funds available for improving Linux!