[Xenomai] Resource utilization

[Xenomai] Resource utilization

[Mathew Benson]

First, I'd like to thank all contributors for Xenomai.  I'm an embedded
software engineer accustomed to working with the more expensive closed
RTOSs.  So far, I'm really happy with Xenomai.

Second, is there a mechanism to monitor resource utilization from user
space?  My application used to measure CPU idle time by busy waiting and
counting in a low priority thread.  Not the best solution, but usually very
portable.  However, the nucleus detects the Linux kernel is getting starved
and flags it as a failure.  I could move this thread to secondary mode, but
I would rather use a native API if one exists.

Similar to the previous question, what other metrics are available?

Re: [Xenomai] Resource utilization

[Gilles Chanteperdrix]

On Tue, Jan 06, 2015 at 10:20:04AM -0600, Mathew Benson wrote:
> First, I'd like to thank all contributors for Xenomai.  I'm an embedded
> software engineer accustomed to working with the more expensive closed
> RTOSs.  So far, I'm really happy with Xenomai.
> 
> Second, is there a mechanism to monitor resource utilization from user
> space?  My application used to measure CPU idle time by busy waiting and
> counting in a low priority thread.  Not the best solution, but usually very
> portable.  However, the nucleus detects the Linux kernel is getting starved
> and flags it as a failure.

You will get the same information with /proc/xenomai/stat


>  I could move this thread to secondary mode, but
> I would rather use a native API if one exists.

That would be another information (the first information is xenomai
load, the second information is the total load). For this
information, your approach is the only available.

> 
> Similar to the previous question, what other metrics are
> available?

other metrics?

-- 
					    Gilles.

Re: [Xenomai] Resource utilization

[Mathew Benson]

Heap utilization, cache faults, page faults, BSS size, thread stack high water mark, context switch count, kernel objects in use, etc...  Thanks.  I forgot about VFS nodes.

Sent from my iPad

> On Jan 7, 2015, at 2:12 AM, Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org> wrote:
> 
>> On Tue, Jan 06, 2015 at 10:20:04AM -0600, Mathew Benson wrote:
>> First, I'd like to thank all contributors for Xenomai.  I'm an embedded
>> software engineer accustomed to working with the more expensive closed
>> RTOSs.  So far, I'm really happy with Xenomai.
>> 
>> Second, is there a mechanism to monitor resource utilization from user
>> space?  My application used to measure CPU idle time by busy waiting and
>> counting in a low priority thread.  Not the best solution, but usually very
>> portable.  However, the nucleus detects the Linux kernel is getting starved
>> and flags it as a failure.
> 
> You will get the same information with /proc/xenomai/stat
> 
> 
>> I could move this thread to secondary mode, but
>> I would rather use a native API if one exists.
> 
> That would be another information (the first information is xenomai
> load, the second information is the total load). For this
> information, your approach is the only available.
> 
>> 
>> Similar to the previous question, what other metrics are
>> available?
> 
> other metrics?
> 
> -- 
>                        Gilles.

Re: [Xenomai] Resource utilization

[Gilles Chanteperdrix]

On Wed, Jan 07, 2015 at 05:06:43AM -0800, Mathew Benson wrote:
> Heap utilization,
> cache faults, page faults, BSS size, thread
> stack high water mark, context switch count, kernel objects in
> use, etc... Thanks. I forgot about VFS nodes.

Xenomai tasks are just Linux tasks, so for all these "metrics", you
should turn to what Linux provides. For the stack high water mark
for instance, I believe neither Linux nor Xenomai have this
information.

-- 
					    Gilles.

Re: [Xenomai] Resource utilization

[Mathew Benson]

Ok.  Thanks.

Sent from my iPhone

> On Jan 7, 2015, at 8:06 AM, Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org> wrote:
> 
>> On Wed, Jan 07, 2015 at 05:06:43AM -0800, Mathew Benson wrote:
>> Heap utilization,
>> cache faults, page faults, BSS size, thread
>> stack high water mark, context switch count, kernel objects in
>> use, etc... Thanks. I forgot about VFS nodes.
> 
> Xenomai tasks are just Linux tasks, so for all these "metrics", you
> should turn to what Linux provides. For the stack high water mark
> for instance, I believe neither Linux nor Xenomai have this
> information.
> 
> -- 
>                        Gilles.

Re: [Xenomai] Resource utilization

[Gilles Chanteperdrix]

On Wed, Jan 07, 2015 at 03:06:51PM +0100, Gilles Chanteperdrix wrote:
> On Wed, Jan 07, 2015 at 05:06:43AM -0800, Mathew Benson wrote:
> > Heap utilization,
> > cache faults, page faults, BSS size, thread
> > stack high water mark, context switch count, kernel objects in
> > use, etc... Thanks. I forgot about VFS nodes.
> 
> Xenomai tasks are just Linux tasks, so for all these "metrics", you
> should turn to what Linux provides. For the stack high water mark
> for instance, I believe neither Linux nor Xenomai have this
> information.

In the case of page faults, and context switch count, mode switches,
you can find this information in /proc/xenomai/sched and
/proc/xenomai/stat

-- 
					    Gilles.

Re: [Xenomai] Resource utilization

[Mathew Benson]

Perfect.  I'll shop around that directory for other interesting data.  Thanks.

Sent from my iPhone

> On Jan 7, 2015, at 10:20 AM, Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org> wrote:
> 
>> On Wed, Jan 07, 2015 at 03:06:51PM +0100, Gilles Chanteperdrix wrote:
>>> On Wed, Jan 07, 2015 at 05:06:43AM -0800, Mathew Benson wrote:
>>> Heap utilization,
>>> cache faults, page faults, BSS size, thread
>>> stack high water mark, context switch count, kernel objects in
>>> use, etc... Thanks. I forgot about VFS nodes.
>> 
>> Xenomai tasks are just Linux tasks, so for all these "metrics", you
>> should turn to what Linux provides. For the stack high water mark
>> for instance, I believe neither Linux nor Xenomai have this
>> information.
> 
> In the case of page faults, and context switch count, mode switches,
> you can find this information in /proc/xenomai/sched and
> /proc/xenomai/stat
> 
> -- 
>                        Gilles.