Re: calibration issue

by the way, I got this result with both vulcano cvs from today and the
rtai-5.0-test2 release with the hal-linux-4.1.18-x86-7.patch .

Jan

On 11/01/2016 12:03 PM, Jan Benda wrote:
> Hi,
>
> I started to test the new rtai-5. The good news is that so far it
> smoothly compiles (inclusively rtai_math) and the tests run. I still
> need to do the kernel config tweaking to be able to say something
> regarding its performance.
>
> If I understand it right from the README, and from removing
> /usr/realtime/calibration/latencies and seeing it appear again,
> calibration is done the first time rtai_sched is insmoded. This is a
> great idea!
>
> However, I was surprised how quickly this happenend, despite what is
> configured in .rtai_config:
>
> CONFIG_RTAI_LATENCY_SELF_CALIBRATION_METRICS="1"
> CONFIG_RTAI_LATENCY_SELF_CALIBRATION_FREQ="10000"
> CONFIG_RTAI_LATENCY_SELF_CALIBRATION_CYCLES="10000"
>
> This tells me that the calibration is done with 10000 samples at 10kHz.
> So it should last 1 second.
>
> Looking into kernel messages I get this after insmodding rati_hal and
> rtai_sched:
>
> [  311.785774] I-pipe: head domain RTAI registered.
> [  311.785795] RTAI[hal]: mounted. ISOL_CPUS_MASK: 0.
> [  311.785797] SYSINFO - # CPUs: 4, TIMER NAME: 'lapic', TIMER IRQ:
> 4353, TIMER FREQ: 90420998, CLOCK NAME: 'tsc', CLOCK FREQ: 2893498000,
> CPU FREQ: 2893498000, LINUX TIMER IRQ: 4353.
> [  323.899915] RTAI[malloc]: global heap size = 2097152 bytes, <BSD>.
> [  323.899951] , kstacks pool size = 524288 bytes.
> [  323.899954] RTAI[sched]: hard timer type/freq = lapic/90420998(Hz);
> timing: oneshot; linear timed lists.
> [  323.899955] RTAI[sched]: Linux timer freq = 1000 (Hz), TimeBase freq
> = 2893498000 hz.
> [  323.899956] RTAI[sched]: timer setup = 114 ns, resched latency = 0 ns.
> [  323.900814] USERMODE CHECK: OK.
> [  323.900817] USERMODE CHECK PROVIDED (ns): KernelLatency -1,
> UserLatency -1.
> [  324.025772] KERNEL SPACE LATENCY ENDED AT CYCLE: 1243, LATENCY =
> 5148, VARIANCE = 33761/10004569, GAIN = 843651/1000000000, LEAST = 2477.
> [  324.102274] AFTER KERNEL CALIBRATION (WITH RTAI TASK, ns):
> KernelLatency 1779, UserLatency -1 (CALIBRATION: PERIOD 10000 (ns), TIME
> 201392037 (ns)).
> [  324.241126] USER SPACE LATENCY ENDED AT CYCLE: 1374, LATENCY =
> 6000.95, VARIANCE = 0.0031114, GAIN = 0.000777849, LEAST = 1980.
> [  324.248263] USERMODE USER SPACE CALIBRATION: OK.
> [  324.248266] AFTER USER CALIBRATION (ns): KernelLatency 1779,
> UserLatency 2073 (CALIBRATION: PERIOD 10000 (ns), TIME 145943905 (ns)).
> [  324.248268] FINAL CALIBRATION SUMMARY (ns): KernelLatency 1779,
> UserLatency 2073.
>
> In line 324.102274 I read that the period was 10000ns, i.e. 100KHz. And
> in the line before the latency test ended after 1243 cycles instead of
> 10000. So no wonder that this was unnoticable quick and not 1 second!
>
> When I configure for 50kHz, I get a period of 50000ns (20kHz). The
> number of cycles did not change, though.
>
> So, CONFIG_RTAI_LATENCY_SELF_CALIBRATION_FREQ is not a frequency, it is
> the period in ns!
>
> And, why does CONFIG_RTAI_LATENCY_SELF_CALIBRATION_CYCLES has no effect
> at all?
>
> I guess this can be fixed easily...
>
> Cheers,
> Jan
>
>
>
> _______________________________________________
> Rtai mailing list
> [email protected]
> https://mail.rtai.org/cgi-bin/mailman/listinfo/rtai

-- 
----------------------------------------------------------------------

Prof. Dr. Jan Benda
Abteilung Neuroethologie
Institut fuer Neurobiologie
Universitaet Tuebingen
Auf der Morgenstelle 28 E                 phone: +49 / 7071 - 29 74573
72076 Tuebingen                             fax: +49 / 7071 - 29 5144

----------------------------------------------------------------------
_______________________________________________
Rtai mailing list
[email protected]
https://mail.rtai.org/cgi-bin/mailman/listinfo/rtai