[Xenomai] Seeking recommendations for SBC with precision timer

[Xenomai] Seeking recommendations for SBC with precision timer

[Howard Lee Harkness]

I have a client requirement for a system that can handle hard real time
with precision timing, and I'm looking for recommendations for a SBC to
support it. Here's what is needed:

1) x86 architecture
2) support for Linux (I will probably be using Xenomai)
3) precision timer with 1 us resolution
4) PC/104 bus
5) lots of I/O (USB 2, I2C, SPI, GPIB, etc)

Anybody here have any experience with a board that has all this?
-- 
Howard Lee Harkness
(214) 390-4896

Re: [Xenomai] Seeking recommendations for SBC with precision timer

[Nicholas Mc Guire]

On Fri, 26 Oct 2012, Howard Lee Harkness wrote:

> I have a client requirement for a system that can handle hard real time
> with precision timing, and I'm looking for recommendations for a SBC to
> support it. Here's what is needed:
> 
> 1) x86 architecture

fanless ?

> 2) support for Linux (I will probably be using Xenomai)
> 3) precision timer with 1 us resolution

many systems will be able to give you a timer reolution of <=1us
but there is no system that will bee able to timstamp an event with
that precission - that is any timestamp you get will have an
uncertenty atleast in the range of a few microseconds to a few
tens of microseconds (thats independent of using Xenomai, Preept-RT
or RTAI)

> 4) PC/104 bus
> 5) lots of I/O (USB 2, I2C, SPI, GPIB, etc)
> 
> Anybody here have any experience with a board that has all this?

it does to some extent depend on the "etc" of the "lots of I/O" unless you
are good with it all being stacked on the PC/104.

you might want to look at the boards in the http:/www.OSADL.org/QA, unfortunately not xenomai based systems in the farm but the jitter/latency numbers will most likely hold for Xenomai based systems with negligable offsets.

thx!
hofrat

Re: [Xenomai] Seeking recommendations for SBC with precision timer

[Howard Lee Harkness]

Thank you for your fast responses!

1) Not necessarily fanless, but that would be nice.

2) Don't need a timestamp. All I need is a precision interval. I want to be
able to send a signal to a device, and read a measurement exactly 11 us (or
5 us, or 17 us, or 31 us, etc.) later, plus or minus no more than about 1
us. <1 us precision would be better, but I think for now 1 us resolution is
sufficient. Reliably, repeatably.

3) The client can stick stuff on the PC/104, but would prefer to have GPIB
and USB2 on-board.

4) I've go some inquiry emails out to a couple of vendors, including
Advantech. I may be getting some evaluation boards Real Soon Now to play
with.

Interesting observation about ICOP boards having a problem with latency
once an hour. Might be able to deal with that by not scheduling a test
during the high-latency period. Also, since the intervals I'm interested in
are short (and there should be nothing else going on), I could sit and spin
while waiting for completion with interrupts off. Although my understanding
is that there are some interrupts you can't turn off.

I'm wondering about the use of a dual core CPU. Can I run one core with
(most) interrupts off, while running the other normally?

Anybody know if Xenomai works with DSL (Damned Small Linux -- a somewhat
unfortunate name)?

All of the real-time stuff I've done to date didn't have an OS at all (I
wrote my own cooperative multitasker), so I'm a bit new at using an RTOS at
this level. I would welcome suggestions for reading material on the subject.

On Fri, Oct 26, 2012 at 7:38 AM, Nicholas Mc Guire der.herr@hofr.at wrote:

> On Fri, 26 Oct 2012, Howard Lee Harkness wrote:
>
> > I have a client requirement for a system that can handle hard real time
> > with precision timing, and I'm looking for recommendations for a SBC to
> > support it. Here's what is needed:
> >
> > 1) x86 architecture
>
> fanless ?
>
> > 2) support for Linux (I will probably be using Xenomai)
> > 3) precision timer with 1 us resolution
>
> many systems will be able to give you a timer reolution of <=1us
> but there is no system that will bee able to timstamp an event with
> that precission - that is any timestamp you get will have an
> uncertenty atleast in the range of a few microseconds to a few
> tens of microseconds (thats independent of using Xenomai, Preept-RT
> or RTAI)
>
> > 4) PC/104 bus
> > 5) lots of I/O (USB 2, I2C, SPI, GPIB, etc)
> >
> > Anybody here have any experience with a board that has all this?
>
> it does to some extent depend on the "etc" of the "lots of I/O" unless you
> are good with it all being stacked on the PC/104.
>
> you might want to look at the boards in the http:/www.OSADL.org/QA,
> unfortunately not xenomai based systems in the farm but the jitter/latency
> numbers will most likely hold for Xenomai based systems with negligable
> offsets.
>
> thx!
> hofrat
>

On Fri, Oct 26, 2012 at 7:38 AM, Giampaolo Bellini <iw2lsi@gmail.com> wrote:

> Hello Howard
>
> I'll suggest you to use an Advantech board... for example a PCM3362 or a
> PCM-9562... as I'm succesfully using these cards with RTAI...
>
> In these years I've test also other boards like ICOP and IEI (IEI Nova
> 9452) but falls in trouble with them... ICOP's boards are very low cost but
> seems to have a hard-coded MTBF routine that results in long latencies
> exactly every hour, IEI Nova maybe a buggy BIOS... not sure...
>
>    Best Regards,
>
>              Giampaolo
>
>
> 2012/10/26 Howard Lee Harkness <howard.lee.harkness@gmail.com>
>
>> I have a client requirement for a system that can handle hard real time
>> with precision timing, and I'm looking for recommendations for a SBC to
>> support it. Here's what is needed:
>>
>> 1) x86 architecture
>> 2) support for Linux (I will probably be using Xenomai)
>> 3) precision timer with 1 us resolution
>> 4) PC/104 bus
>> 5) lots of I/O (USB 2, I2C, SPI, GPIB, etc)
>>
>> Anybody here have any experience with a board that has all this?
>> --
>> Howard Lee Harkness
>> (214) 390-4896
>> _______________________________________________
>> Xenomai mailing list
>> Xenomai@xenomai.org
>> http://www.xenomai.org/mailman/listinfo/xenomai
>>
>
>

-- 
Howard Lee Harkness
(214) 390-4896

Re: [Xenomai] Seeking recommendations for SBC with precision timer

[Gilles Chanteperdrix]

On 10/26/2012 04:46 PM, Howard Lee Harkness wrote:
> Thank you for your fast responses!
> 
> 1) Not necessarily fanless, but that would be nice.
> 
> 2) Don't need a timestamp. All I need is a precision interval. I want to be
> able to send a signal to a device, and read a measurement exactly 11 us (or
> 5 us, or 17 us, or 31 us, etc.) later, plus or minus no more than about 1
> us. <1 us precision would be better, but I think for now 1 us resolution is
> sufficient. Reliably, repeatably.

What you are asking is a latency of less than 1us, not a resolution.
Here is a graph showing the (user-space wake-up) latencies we get with
various configurations of Xenomai 2.6.1 over Linux 3.4.6 on an atom 230:

http://xenomai.org/~gch/core-3.4-latencies/atom.png

These are user-space wake-up latencies, you get better result with irqs
handlers. But 1us, that is a bit hard. The only way I know to do this,
is, for instance, if you know that the maximum irq latency is 15us,
wake-up 15us before the wanted wake-up date, and spin in the timer
handler until the time you want, using a clocksource which has a better
resolution than 1us. x86 generally use the tsc which is such a clocksource.


-- 
					    Gilles.

Re: [Xenomai] Seeking recommendations for SBC with precision timer

[Howard Lee Harkness]

Ah, that looks like it might be the solution. Need to dig a bit deeper on
this subject.

Back last time I was doing this sort of thing, I was using a 286, which
didn't have a TSC. Plus I only had to do about 500 us resolution on the
timings. I contracted with a friend to do a simple countdown timer for the
PC bus, which did the trick. That was more than 15 years ago. The
semiconductor testing business has gotten a little more complex since then.

On Fri, Oct 26, 2012 at 10:11 AM, Gilles Chanteperdrix <
gilles.chanteperdrix@xenomai.org> wrote:

> The only way I know to do this,
> is, for instance, if you know that the maximum irq latency is 15us,
> wake-up 15us before the wanted wake-up date, and spin in the timer
> handler until the time you want, using a clocksource which has a better
> resolution than 1us. x86 generally use the tsc which is such a clocksource.
>



-- 
Howard Lee Harkness
(214) 390-4896

Re: [Xenomai] Seeking recommendations for SBC with precision timer

[Gilles Chanteperdrix]

On 10/26/2012 06:34 PM, Howard Lee Harkness wrote:
> Ah, that looks like it might be the solution. Need to dig a bit deeper on
> this subject.

If you use the the RTDM API, you would use the rtdm_task_busy_sleep 
routine:
http://www.xenomai.org/documentation/xenomai-2.6/html/api/group__rtdmtask.html#gaa3940cfff7cc72d7bc064a3e279d74ac

> 
> Back last time I was doing this sort of thing, I was using a 286, which
> didn't have a TSC. Plus I only had to do about 500 us resolution on the
> timings. I contracted with a friend to do a simple countdown timer for the
> PC bus, which did the trick. That was more than 15 years ago. The
> semiconductor testing business has gotten a little more complex since then.

I have to admit that when I had a 286, I was still in high-school
and programming it in Turbo Pascal for plotting function graphs
for my math assignments...

-- 
					    Gilles.