[Xenomai-core] ... the rt-preempt patch once been in fusion

[Xenomai-core] ... the rt-preempt patch once been in fusion

[Hannes Mayer]

Hi all!

Well, subject says almost all.
I remember there has been an effort to make a fusion/rt-preempt
patch conglomerate back then.
I'm just curious (as I've just been toying with the Molnar
patch) - has that thing been dropped in xeno completely ?

Thanks and best regards,
Hannes.

Re: [Xenomai-core] ... the rt-preempt patch once been in fusion

[Philippe Gerum]

Hannes Mayer wrote:
> Hi all!
> 
> Well, subject says almost all.
> I remember there has been an effort to make a fusion/rt-preempt
> patch conglomerate back then.
> I'm just curious (as I've just been toying with the Molnar
> patch) - has that thing been dropped in xeno completely ?
> 

Clearly not, it is still on the project's agenda, but PREEMPT_RT has been too much 
of a moving target recently, not to speak of the vanilla 2.6 kernel itself. Even 
if Xeno's core has not changed in the same proportions, a large refactoring has 
taken place for 2.1 though.

Additionally, PREEMPT_RT's infrastructure has evolved a lot since Xeno was first 
ported over such kernel, and the way to interface both sub-systems likely needs to 
be re-thought. This combo cannot be done overnight, but I think that we are 
getting close to point where it would be possible to rework on this issue.

This work is likely going to start by a test campaign evaluating PREEMPT_RT's and 
Xeno's nucleus performances in a series of simple and relevant test cases over 
different ranges of platforms, so that we could measure up to which point (and 
under which condition) the real-time spectrum would be widened by using a 
combination of both.

-- 

Philippe.

Re: [Xenomai-core] ... the rt-preempt patch once been in fusion

[Hannes Mayer]

Ciao Philippe!

Philippe Gerum wrote:
[...]
> Clearly not, it is still on the project's agenda, but PREEMPT_RT has 
> been too much of a moving target recently, not to speak of the vanilla 
> 2.6 kernel itself. 

Recently ? As for Kernel 2.6 there has always been too much
development for a stable release IMO, so I still prefer 2.4,
but that's a solely personal opinion. On the other hand, 2.6
has been quite stable for me (desktop and devel-stress-testing).

> This combo cannot be done 
> overnight, 

I was not complaining - just curious :-)

> This work is likely going to start by a test campaign evaluating 
> PREEMPT_RT's and Xeno's nucleus performances in a series of simple and 
> relevant test cases over different ranges of platforms, so that we could 

Testing ?
FWIW, I've just completed another page about the vanilla vs.
preempt_rt thing:
http://www.captain.at/howto-linux-real-time-patch.php
If you have the one or another additional stress test, I'd be more
than happy to add it to the page.

Thanks and best regards,
Hannes.

Re: [Xenomai-core] ... the rt-preempt patch once been in fusion

[Philippe Gerum]

Hannes Mayer wrote:
> Ciao Philippe!
> 
> Philippe Gerum wrote:
> [...]
> 
>> Clearly not, it is still on the project's agenda, but PREEMPT_RT has 
>> been too much of a moving target recently, not to speak of the vanilla 
>> 2.6 kernel itself. 
> 
> 
> Recently ? As for Kernel 2.6 there has always been too much
> development for a stable release IMO, so I still prefer 2.4,
> but that's a solely personal opinion. On the other hand, 2.6
> has been quite stable for me (desktop and devel-stress-testing).
> 

I've not talked about instability for 2.6, but rather significant changes in some 
low-level layers. Incidentally, some of them were triggered by portion of the 
PREEMPT_RT infrastructure leaking to the vanilla kernel. Unfortunately, this 
sometimes causes issues you have to deal with when adapting Adeos to each new release.

>> This combo cannot be done overnight, 
> 
> 
> I was not complaining - just curious :-)
> 
>> This work is likely going to start by a test campaign evaluating 
>> PREEMPT_RT's and Xeno's nucleus performances in a series of simple and 
>> relevant test cases over different ranges of platforms, so that we could 
> 
> 
> Testing ?
> FWIW, I've just completed another page about the vanilla vs.
> preempt_rt thing:
> http://www.captain.at/howto-linux-real-time-patch.php
> If you have the one or another additional stress test, I'd be more
> than happy to add it to the page.
> 

Ack. Thanks.

> Thanks and best regards,
> Hannes.
> 


-- 

Philippe.