Xenomai vs RTAI for real-time performance

Hello,

I just saw a post in "ROS Answers" which mentions that Xenomai should be used as a real-time target if one wants to have real-time performance below 5ms. Is this the case? In general, are Xenomai and RTAI comparable in terms of performance from the point of view of Orocos?

Thanks,
JML

Jason Michel Lambert
Agent de Recherche/Research Officer
Centre de Technologies de Fabrication Aérospatiale/
Aerospace Manufacturing Technology Center
Conseil National de Recherche du Canada/
National Research Council Canada
5145 Avenue Decelles
Campus de l'Université de Montréal
Montréal, Quebec, H3T-2B2
Tel: 514-283-9175
Fax: 514-283-9604
<mailto:michel [dot] lambert [..] ...>

Xenomai vs RTAI for real-time performance

2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
> Hello,
>
>
>
> I just saw a post in “ROS Answers” which mentions that Xenomai should be
> used as a real-time target if one wants to have real-time performance below
> 5ms. Is this the case?

I'm not sure ROS communication allows such low performances. I
personnaly never use ROS under 100ms but It's not in RT context.

> In general, are Xenomai and RTAI comparable in terms
> of performance from the point of view of Orocos?
>

You can have really low cycle with Xenomai+Orocos. I acheive 10ms
without difficulites, but they are working exemples below that.

>
>
> Thanks,
>
> JML
>
>
>
> Jason Michel Lambert
> Agent de Recherche/Research Officer
> Centre de Technologies de Fabrication Aérospatiale/
> Aerospace Manufacturing Technology Center
> Conseil National de Recherche du Canada/
> National Research Council Canada
> 5145 Avenue Decelles
> Campus de l'Université de Montréal
> Montréal, Quebec, H3T-2B2
> Tel: 514-283-9175
> Fax: 514-283-9604
> <mailto:michel [dot] lambert [..] ...>
>
>
>
>
> --
> Orocos-Users mailing list
> Orocos-Users [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
>

Xenomai vs RTAI for real-time performance

2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
> Hello,
>
>
>
> I just saw a post in "ROS Answers" which mentions that Xenomai should
> be used as a real-time target if one wants to have real-time
> performance below 5ms. Is this the case?

I'm not sure ROS communication allows such low performances. I personnaly never use ROS under 100ms but It's not in RT context.

Sorry I should've been more specific. The 5ms did not involve ROS, just Orocos.

> In general, are Xenomai and RTAI comparable in terms of performance
> from the point of view of Orocos?
>

You can have really low cycle with Xenomai+Orocos. I acheive 10ms without difficulites, but they are working exemples below that.

Has anyone worked in the 1ms range using Orocos and RTAI? I guess I'm trying to convince myself on which real-time framework to use, I'm under the impression that RTAI has better hardware support through Comedi, but it seems lots of Orocos people are using Xenomai. Is there a particular reason for this?

>
>
> Thanks,
>
> JML
>
>
>
> Jason Michel Lambert
> Agent de Recherche/Research Officer
> Centre de Technologies de Fabrication Aérospatiale/ Aerospace
> Manufacturing Technology Center Conseil National de Recherche du
> Canada/ National Research Council Canada
> 5145 Avenue Decelles
> Campus de l'Université de Montréal
> Montréal, Quebec, H3T-2B2
> Tel: 514-283-9175
> Fax: 514-283-9604
> <mailto:michel [dot] lambert [..] ...>
>
>
>
>
> --
> Orocos-Users mailing list
> Orocos-Users [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
>

Xenomai vs RTAI for real-time performance

On Mon, Feb 04, 2013 at 02:32:33PM -0500, Lambert, Michel wrote:
>
> 2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
> > Hello,
> >
> >
> >
> > I just saw a post in "ROS Answers" which mentions that Xenomai should
> > be used as a real-time target if one wants to have real-time
> > performance below 5ms. Is this the case?
>
> I'm not sure ROS communication allows such low performances. I personnaly never use ROS under 100ms but It's not in RT context.
>
> Sorry I should've been more specific. The 5ms did not involve ROS, just Orocos.
>
> > In general, are Xenomai and RTAI comparable in terms of performance
> > from the point of view of Orocos?
> >
>
> You can have really low cycle with Xenomai+Orocos. I acheive 10ms without difficulites, but they are working exemples below that.
>
> Has anyone worked in the 1ms range using Orocos and RTAI? I guess
> I'm trying to convince myself on which real-time framework to use,
> I'm under the impression that RTAI has better hardware support
> through Comedi, but it seems lots of Orocos people are using
> Xenomai. Is there a particular reason for this?

Yes, Xenomain has the cleaner, more maintainable codebase and better
portability through skins. The performance of RTAI might be a little
better, however this comes at the price of maintainability. With
Xenomai you can use Analogy which is the derived from Comedi/RTDM.

Regards
Markus

Xenomai vs RTAI for real-time performance

> -----Original Message-----
> From: Markus Klotzbuecher
> [mailto:markus [dot] klotzbuecher [..] ...]
> Sent: February-05-13 9:23 AM
> To: Lambert, Michel
> Cc: Willy Lambert; orocos-users [..] ...
> Subject: Re: [Orocos-users] Xenomai vs RTAI for real-time performance
>
> On Mon, Feb 04, 2013 at 02:32:33PM -0500, Lambert, Michel wrote:
> >
> > 2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
> > > Hello,
> > >
> > >
> > >
> > > I just saw a post in "ROS Answers" which mentions that Xenomai
> > > should be used as a real-time target if one wants to have real-time
> > > performance below 5ms. Is this the case?
> >
> > I'm not sure ROS communication allows such low performances. I
> personnaly never use ROS under 100ms but It's not in RT context.
> >
> > Sorry I should've been more specific. The 5ms did not involve ROS, just
> Orocos.
> >
> > > In general, are Xenomai and RTAI comparable in terms of performance
> > > from the point of view of Orocos?
> > >
> >
> > You can have really low cycle with Xenomai+Orocos. I acheive 10ms without
> difficulites, but they are working exemples below that.
> >
> > Has anyone worked in the 1ms range using Orocos and RTAI? I guess I'm
> > trying to convince myself on which real-time framework to use, I'm
> > under the impression that RTAI has better hardware support through
> > Comedi, but it seems lots of Orocos people are using Xenomai. Is there
> > a particular reason for this?
>
> Yes, Xenomain has the cleaner, more maintainable codebase and better
> portability through skins. The performance of RTAI might be a little better,
> however this comes at the price of maintainability. With Xenomai you can
> use Analogy which is the derived from Comedi/RTDM.

O.k., looks like Xenomai is a safe(r)(est) bet. Thanks for the comments!
>
> Regards
> Markus

Xenomai vs RTAI for real-time performance

On Feb 4, 2013, at 14:32 , Lambert, Michel wrote:

>
>
> 2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
>> Hello,
>>
>>
>>
>> I just saw a post in "ROS Answers" which mentions that Xenomai should
>> be used as a real-time target if one wants to have real-time
>> performance below 5ms. Is this the case?
>
> I'm not sure ROS communication allows such low performances. I personnaly never use ROS under 100ms but It's not in RT context.
>
> Sorry I should've been more specific. The 5ms did not involve ROS, just Orocos.
>
>> In general, are Xenomai and RTAI comparable in terms of performance
>> from the point of view of Orocos?
>>
>
> You can have really low cycle with Xenomai+Orocos. I acheive 10ms without difficulites, but they are working exemples below that.
>
> Has anyone worked in the 1ms range using Orocos and RTAI? I guess I'm trying to convince myself on which real-time framework to use, I'm under the impression that RTAI has better hardware support through Comedi, but it seems lots of Orocos people are using Xenomai. Is there a particular reason for this?

We use Orocos plus the PREEMPT_RT patch to run at 2ms. Standard desktop hardware easily supports this.
S

Xenomai vs RTAI for real-time performance

On Tue, Feb 5, 2013 at 12:04 AM, S Roderick <kiwi [dot] net [..] ...> wrote:

> On Feb 4, 2013, at 14:32 , Lambert, Michel wrote:
>
> >
> >
> > 2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
> >> Hello,
> >>
> >>
> >>
> >> I just saw a post in "ROS Answers" which mentions that Xenomai should
> >> be used as a real-time target if one wants to have real-time
> >> performance below 5ms. Is this the case?
> >
> > I'm not sure ROS communication allows such low performances. I
> personnaly never use ROS under 100ms but It's not in RT context.
> >
> > Sorry I should've been more specific. The 5ms did not involve ROS, just
> Orocos.
> >
> >> In general, are Xenomai and RTAI comparable in terms of performance
> >> from the point of view of Orocos?
> >>
> >
> > You can have really low cycle with Xenomai+Orocos. I acheive 10ms
> without difficulites, but they are working exemples below that.
> >
> > Has anyone worked in the 1ms range using Orocos and RTAI? I guess I'm
> trying to convince myself on which real-time framework to use, I'm under
> the impression that RTAI has better hardware support through Comedi, but it
> seems lots of Orocos people are using Xenomai. Is there a particular reason
> for this?
>
> We use Orocos plus the PREEMPT_RT patch to run at 2ms. Standard desktop
> hardware easily supports this.
> S
>
I have been using Orocos with the PREEMPT_RT patch too, running at 0.5 ms
on a standard desktop without any noticeable problem.

Bert

>
> --
> Orocos-Users mailing list
> Orocos-Users [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
>

Xenomai vs RTAI for real-time performance

2013/2/5 Bert Willaert <bert [dot] willaert [..] ...>:
>
>
> On Tue, Feb 5, 2013 at 12:04 AM, S Roderick <kiwi [dot] net [..] ...> wrote:
>>
>> On Feb 4, 2013, at 14:32 , Lambert, Michel wrote:
>>
>> >
>> >
>> > 2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
>> >> Hello,
>> >>
>> >>
>> >>
>> >> I just saw a post in "ROS Answers" which mentions that Xenomai should
>> >> be used as a real-time target if one wants to have real-time
>> >> performance below 5ms. Is this the case?
>> >
>> > I'm not sure ROS communication allows such low performances. I
>> > personnaly never use ROS under 100ms but It's not in RT context.
>> >
>> > Sorry I should've been more specific. The 5ms did not involve ROS, just
>> > Orocos.
>> >
>> >> In general, are Xenomai and RTAI comparable in terms of performance
>> >> from the point of view of Orocos?
>> >>
>> >
>> > You can have really low cycle with Xenomai+Orocos. I acheive 10ms
>> > without difficulites, but they are working exemples below that.
>> >
>> > Has anyone worked in the 1ms range using Orocos and RTAI? I guess I'm
>> > trying to convince myself on which real-time framework to use, I'm under the
>> > impression that RTAI has better hardware support through Comedi, but it
>> > seems lots of Orocos people are using Xenomai. Is there a particular reason
>> > for this?
>>
>> We use Orocos plus the PREEMPT_RT patch to run at 2ms. Standard desktop
>> hardware easily supports this.
>> S
>
> I have been using Orocos with the PREEMPT_RT patch too, running at 0.5 ms on
> a standard desktop without any noticeable problem.
>
> Bert

I'm really surprised that you can acheive so little cycle time with a
"standard desktop" just with PREEMP_RT.

When you say "without any noticeable problem", do you mean that "in
general (with a high probability) I have 0.5ms" or "I never have
orverheads" ? What are you mean to check this times ?

Setting up an RT OS is some work. It would be magic if you just had to
select one option in the kernel. Or does that mean that xenomai is
required for stuff running faster that this (like 10 micro seconds)

>>
>>
>> --
>> Orocos-Users mailing list
>> Orocos-Users [..] ...
>> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
>
>
>
> --
> Orocos-Users mailing list
> Orocos-Users [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
>

Xenomai vs RTAI for real-time performance

On Feb 5, 2013, at 04:41 , Willy Lambert wrote:

> 2013/2/5 Bert Willaert <bert [dot] willaert [..] ...>:
>>
>>
>> On Tue, Feb 5, 2013 at 12:04 AM, S Roderick <kiwi [dot] net [..] ...> wrote:
>>>
>>> On Feb 4, 2013, at 14:32 , Lambert, Michel wrote:
>>>
>>>>
>>>>
>>>> 2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
>>>>> Hello,
>>>>>
>>>>>
>>>>>
>>>>> I just saw a post in "ROS Answers" which mentions that Xenomai should
>>>>> be used as a real-time target if one wants to have real-time
>>>>> performance below 5ms. Is this the case?
>>>>
>>>> I'm not sure ROS communication allows such low performances. I
>>>> personnaly never use ROS under 100ms but It's not in RT context.
>>>>
>>>> Sorry I should've been more specific. The 5ms did not involve ROS, just
>>>> Orocos.
>>>>
>>>>> In general, are Xenomai and RTAI comparable in terms of performance
>>>>> from the point of view of Orocos?
>>>>>
>>>>
>>>> You can have really low cycle with Xenomai+Orocos. I acheive 10ms
>>>> without difficulites, but they are working exemples below that.
>>>>
>>>> Has anyone worked in the 1ms range using Orocos and RTAI? I guess I'm
>>>> trying to convince myself on which real-time framework to use, I'm under the
>>>> impression that RTAI has better hardware support through Comedi, but it
>>>> seems lots of Orocos people are using Xenomai. Is there a particular reason
>>>> for this?
>>>
>>> We use Orocos plus the PREEMPT_RT patch to run at 2ms. Standard desktop
>>> hardware easily supports this.
>>> S
>>
>> I have been using Orocos with the PREEMPT_RT patch too, running at 0.5 ms on
>> a standard desktop without any noticeable problem.
>>
>> Bert
>
> I'm really surprised that you can acheive so little cycle time with a
> "standard desktop" just with PREEMP_RT.

Why? The RT developers regularly post jitter numbers in the 10's of microseconds range. That's about what we see. Check out osadl ...

> When you say "without any noticeable problem", do you mean that "in
> general (with a high probability) I have 0.5ms" or "I never have
> orverheads" ? What are you mean to check this times ?

Our system always collects jitter/latency measurements. They're usually in the 10's of microseconds range, on an off the shelf Dell dekstop. There are certainly plenty of possible causes of issues (e.g. SMI), which are fairly well documented on the Linux RT wiki. I'm not saying that we don't have a probability of issue; but if we do, it's of the once a day variety.

> Setting up an RT OS is some work. It would be magic if you just had to
> select one option in the kernel. Or does that mean that xenomai is
> required for stuff running faster that this (like 10 micro seconds)

Actually, IMHO setting up the OS isn't such an issue (except for Xenomai - it's so tied to the hardware). Setting up the application to remain RT is more trouble.
S

Xenomai vs RTAI for real-time performance

My personal experience is that I can comfortably achieve loops of 1ms with
Xenomai.
I have no experience with RTAI.

Davide

On Mon, Feb 4, 2013 at 8:32 PM, Lambert, Michel <
JasonMichel [dot] Lambert [..] ...> wrote:

>
>
> 2013/2/4 Lambert, Michel <JasonMichel [dot] Lambert [..] ...>:
> > Hello,
> >
> >
> >
> > I just saw a post in "ROS Answers" which mentions that Xenomai should
> > be used as a real-time target if one wants to have real-time
> > performance below 5ms. Is this the case?
>
> I'm not sure ROS communication allows such low performances. I personnaly
> never use ROS under 100ms but It's not in RT context.
>
> Sorry I should've been more specific. The 5ms did not involve ROS, just
> Orocos.
>
> > In general, are Xenomai and RTAI comparable in terms of performance
> > from the point of view of Orocos?
> >
>
> You can have really low cycle with Xenomai+Orocos. I acheive 10ms without
> difficulites, but they are working exemples below that.
>
> Has anyone worked in the 1ms range using Orocos and RTAI? I guess I'm
> trying to convince myself on which real-time framework to use, I'm under
> the impression that RTAI has better hardware support through Comedi, but it
> seems lots of Orocos people are using Xenomai. Is there a particular reason
> for this?
>
> >
> >
> > Thanks,
> >
> > JML
> >
> >
> >
> > Jason Michel Lambert
> > Agent de Recherche/Research Officer
> > Centre de Technologies de Fabrication Aérospatiale/ Aerospace
> > Manufacturing Technology Center Conseil National de Recherche du
> > Canada/ National Research Council Canada
> > 5145 Avenue Decelles
> > Campus de l'Université de Montréal
> > Montréal, Quebec, H3T-2B2
> > Tel: 514-283-9175
> > Fax: 514-283-9604
> > <mailto:michel [dot] lambert [..] ...>
> >
> >
> >
> >
> > --
> > Orocos-Users mailing list
> > Orocos-Users [..] ...
> > http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
> >
> --
> Orocos-Users mailing list
> Orocos-Users [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
>