* [Xenomai-help] negative values of latency/klatency
@ 2006-02-27 16:12 Rudolf Marek
2006-02-28 12:33 ` Jan Kiszka
0 siblings, 1 reply; 6+ messages in thread
From: Rudolf Marek @ 2006-02-27 16:12 UTC (permalink / raw)
To: xenomai
Hello all,
I tried the latency and klatency demo on some PPC procesor(s) (e500 and 60x) and the measurement results are rather strange.
I observed this with arbitrary version of Adeos/ipipe and on Xenomai 2.0.3 older 2.0.x too.
-sh-3.00# ./run -- -T 10 -h -l 5
...
RTT| 00:00:01
RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
RTD| -4525| -4323| 5777| 0| -4525| 5777
RTD| -4525| -4444| -1777| 0| -4525| 5777
RTD| -4525| -4444| -565| 0| -4525| 5777
RTD| -4525| -4444| -1656| 0| -4525| 5777
RTD| -4525| -4444| -1979| 0| -4525| 5777
RTT| 00:00:06
RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
RTD| -4525| -4444| -2020| 0| -4525| 5777
RTD| -4525| -4444| -1050| 0| -4525| 5777
RTD| -4525| -4444| -1616| 0| -4525| 5777
RTD| -4525| -4444| -1898| 0| -4525| 5777
RTD| -4525| -4444| -1939| 0| -4525| 5777
...
Well if I set scheduling latency from automatic "0" to "1" I get:
-sh-3.00# ./run -- -T 10 -h -l 5
RTT| 00:00:01
RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
RTD| 4767| 4929| 15191| 0| 4767| 15191
RTD| 4767| 4808| 8282| 0| 4767| 15191
RTD| 4767| 4808| 8080| 0| 4767| 15191
RTD| 4808| 4808| 7838| 0| 4767| 15191
RTD| 4767| 4808| 7272| 0| 4767| 15191
RTT| 00:00:06
RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
RTD| 4808| 4808| 7555| 0| 4767| 15191
RTD| 4767| 4808| 7959| 0| 4767| 15191
RTD| 4767| 4808| 7393| 0| 4767| 15191
RTD| 4808| 4808| 7191| 0| 4767| 15191
RTD| 4767| 4808| 7313| 0| 4767| 15191
Is this a bug or feature please? Can someone throw the light?
Good would be to print the units to the numbers too (ns).
Thank you,
Regards,
Rudolf
chipset : 8541E
Vendor : Freescale Semiconductor
Machine : CDS - MPC8541E (11)
clock : 396MHz
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [Xenomai-help] negative values of latency/klatency
2006-02-27 16:12 [Xenomai-help] negative values of latency/klatency Rudolf Marek
@ 2006-02-28 12:33 ` Jan Kiszka
2006-02-28 12:52 ` Philippe Gerum
2006-02-28 14:13 ` [Xenomai-core] " Jan Kiszka
0 siblings, 2 replies; 6+ messages in thread
From: Jan Kiszka @ 2006-02-28 12:33 UTC (permalink / raw)
To: Rudolf Marek; +Cc: xenomai
[-- Attachment #1: Type: text/plain, Size: 3023 bytes --]
Rudolf Marek wrote:
> Hello all,
>
> I tried the latency and klatency demo on some PPC procesor(s) (e500 and 60x) and the measurement results are rather strange.
> I observed this with arbitrary version of Adeos/ipipe and on Xenomai 2.0.3 older 2.0.x too.
>
> -sh-3.00# ./run -- -T 10 -h -l 5
> ...
> RTT| 00:00:01
> RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
> RTD| -4525| -4323| 5777| 0| -4525| 5777
> RTD| -4525| -4444| -1777| 0| -4525| 5777
> RTD| -4525| -4444| -565| 0| -4525| 5777
> RTD| -4525| -4444| -1656| 0| -4525| 5777
> RTD| -4525| -4444| -1979| 0| -4525| 5777
> RTT| 00:00:06
> RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
> RTD| -4525| -4444| -2020| 0| -4525| 5777
> RTD| -4525| -4444| -1050| 0| -4525| 5777
> RTD| -4525| -4444| -1616| 0| -4525| 5777
> RTD| -4525| -4444| -1898| 0| -4525| 5777
> RTD| -4525| -4444| -1939| 0| -4525| 5777
> ...
> Well if I set scheduling latency from automatic "0" to "1" I get:
0 means automatic latency compensation (which does not seem to work that
well here, it overcompensates), while 1 effectively means "no
hardware/software scheduling latency" (=> 1 ns). See also help of the
related kernel configuration parameters.
>
> -sh-3.00# ./run -- -T 10 -h -l 5
>
> RTT| 00:00:01
> RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
> RTD| 4767| 4929| 15191| 0| 4767| 15191
> RTD| 4767| 4808| 8282| 0| 4767| 15191
> RTD| 4767| 4808| 8080| 0| 4767| 15191
> RTD| 4808| 4808| 7838| 0| 4767| 15191
> RTD| 4767| 4808| 7272| 0| 4767| 15191
> RTT| 00:00:06
> RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
> RTD| 4808| 4808| 7555| 0| 4767| 15191
> RTD| 4767| 4808| 7959| 0| 4767| 15191
> RTD| 4767| 4808| 7393| 0| 4767| 15191
> RTD| 4808| 4808| 7191| 0| 4767| 15191
> RTD| 4767| 4808| 7313| 0| 4767| 15191
>
> Is this a bug or feature please? Can someone throw the light?
> Good would be to print the units to the numbers too (ns).
>
That was likely a layout question of the latency tool's output. We could
simply dump something like "All latencies in nanoseconds" during
start-up. Would this be more helpful?
Jan
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [Xenomai-help] negative values of latency/klatency
2006-02-28 12:33 ` Jan Kiszka
@ 2006-02-28 12:52 ` Philippe Gerum
2006-02-28 13:02 ` Philippe Gerum
2006-02-28 14:13 ` [Xenomai-core] " Jan Kiszka
1 sibling, 1 reply; 6+ messages in thread
From: Philippe Gerum @ 2006-02-28 12:52 UTC (permalink / raw)
To: Jan Kiszka; +Cc: xenomai, Rudolf Marek
Jan Kiszka wrote:
> Rudolf Marek wrote:
>
>>Hello all,
>>
>>I tried the latency and klatency demo on some PPC procesor(s) (e500 and 60x) and the measurement results are rather strange.
>>I observed this with arbitrary version of Adeos/ipipe and on Xenomai 2.0.3 older 2.0.x too.
>>
>>-sh-3.00# ./run -- -T 10 -h -l 5
>>...
>>RTT| 00:00:01
>>RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>>RTD| -4525| -4323| 5777| 0| -4525| 5777
>>RTD| -4525| -4444| -1777| 0| -4525| 5777
>>RTD| -4525| -4444| -565| 0| -4525| 5777
>>RTD| -4525| -4444| -1656| 0| -4525| 5777
>>RTD| -4525| -4444| -1979| 0| -4525| 5777
>>RTT| 00:00:06
>>RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>>RTD| -4525| -4444| -2020| 0| -4525| 5777
>>RTD| -4525| -4444| -1050| 0| -4525| 5777
>>RTD| -4525| -4444| -1616| 0| -4525| 5777
>>RTD| -4525| -4444| -1898| 0| -4525| 5777
>>RTD| -4525| -4444| -1939| 0| -4525| 5777
>>...
>>Well if I set scheduling latency from automatic "0" to "1" I get:
>
>
> 0 means automatic latency compensation (which does not seem to work that
> well here, it overcompensates),
Dynamic determination of the intrinsic latency during nucleus startup has been
dropped some time ago (circa fusion 0.6.8), because it just did not provide
accurate results. Since then, when 0 is specified in Kconfig, the value read from
include/asm-*/calibration.h is substituted. In this case, 9500 ns are taken for a
ppc32 platform, which is likely too high for this board. Setting ~4500 in Kconfig
would better approximate the latency.
while 1 effectively means "no
> hardware/software scheduling latency" (=> 1 ns). See also help of the
> related kernel configuration parameters.
>
>
>>-sh-3.00# ./run -- -T 10 -h -l 5
>>
>>RTT| 00:00:01
>>RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>>RTD| 4767| 4929| 15191| 0| 4767| 15191
>>RTD| 4767| 4808| 8282| 0| 4767| 15191
>>RTD| 4767| 4808| 8080| 0| 4767| 15191
>>RTD| 4808| 4808| 7838| 0| 4767| 15191
>>RTD| 4767| 4808| 7272| 0| 4767| 15191
>>RTT| 00:00:06
>>RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>>RTD| 4808| 4808| 7555| 0| 4767| 15191
>>RTD| 4767| 4808| 7959| 0| 4767| 15191
>>RTD| 4767| 4808| 7393| 0| 4767| 15191
>>RTD| 4808| 4808| 7191| 0| 4767| 15191
>>RTD| 4767| 4808| 7313| 0| 4767| 15191
>>
>>Is this a bug or feature please? Can someone throw the light?
>>Good would be to print the units to the numbers too (ns).
>>
>
>
> That was likely a layout question of the latency tool's output. We could
> simply dump something like "All latencies in nanoseconds" during
> start-up. Would this be more helpful?
>
> Jan
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help
--
Philippe.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [Xenomai-help] negative values of latency/klatency
2006-02-28 12:52 ` Philippe Gerum
@ 2006-02-28 13:02 ` Philippe Gerum
0 siblings, 0 replies; 6+ messages in thread
From: Philippe Gerum @ 2006-02-28 13:02 UTC (permalink / raw)
To: Philippe Gerum; +Cc: xenomai, Rudolf Marek, Jan Kiszka
Philippe Gerum wrote:
> Jan Kiszka wrote:
>
>> Rudolf Marek wrote:
>>
>>> Hello all,
>>>
>>> I tried the latency and klatency demo on some PPC procesor(s) (e500
>>> and 60x) and the measurement results are rather strange.
>>> I observed this with arbitrary version of Adeos/ipipe and on Xenomai
>>> 2.0.3 older 2.0.x too.
>>>
>>> -sh-3.00# ./run -- -T 10 -h -l 5
>>> ...
>>> RTT| 00:00:01
>>> RTH|----klat min|----klat avg|----klat max| overrun|---klat
>>> best|--klat worst
>>> RTD| -4525| -4323| 5777| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -1777| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -565| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -1656| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -1979| 0|
>>> -4525| 5777
>>> RTT| 00:00:06
>>> RTH|----klat min|----klat avg|----klat max| overrun|---klat
>>> best|--klat worst
>>> RTD| -4525| -4444| -2020| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -1050| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -1616| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -1898| 0|
>>> -4525| 5777
>>> RTD| -4525| -4444| -1939| 0|
>>> -4525| 5777
>>> ...
>>> Well if I set scheduling latency from automatic "0" to "1" I get:
>>
>>
>>
>> 0 means automatic latency compensation (which does not seem to work that
>> well here, it overcompensates),
>
>
> Dynamic determination of the intrinsic latency during nucleus startup
> has been dropped some time ago (circa fusion 0.6.8), because it just did
> not provide accurate results. Since then, when 0 is specified in
> Kconfig, the value read from include/asm-*/calibration.h is substituted.
> In this case, 9500 ns are taken for a ppc32 platform, which is likely
> too high for this board. Setting ~4500 in Kconfig would better
> approximate the latency.
>
Btw, the following would work too in order to easily test various settings
dynamically:
$ echo 4500 > /proc/xenomai/latency
> while 1 effectively means "no
>
>> hardware/software scheduling latency" (=> 1 ns). See also help of the
>> related kernel configuration parameters.
>>
>>
>>> -sh-3.00# ./run -- -T 10 -h -l 5
>>>
>>> RTT| 00:00:01
>>> RTH|----klat min|----klat avg|----klat max| overrun|---klat
>>> best|--klat worst
>>> RTD| 4767| 4929| 15191| 0|
>>> 4767| 15191
>>> RTD| 4767| 4808| 8282| 0|
>>> 4767| 15191
>>> RTD| 4767| 4808| 8080| 0|
>>> 4767| 15191
>>> RTD| 4808| 4808| 7838| 0|
>>> 4767| 15191
>>> RTD| 4767| 4808| 7272| 0|
>>> 4767| 15191
>>> RTT| 00:00:06
>>> RTH|----klat min|----klat avg|----klat max| overrun|---klat
>>> best|--klat worst
>>> RTD| 4808| 4808| 7555| 0|
>>> 4767| 15191
>>> RTD| 4767| 4808| 7959| 0|
>>> 4767| 15191
>>> RTD| 4767| 4808| 7393| 0|
>>> 4767| 15191
>>> RTD| 4808| 4808| 7191| 0|
>>> 4767| 15191
>>> RTD| 4767| 4808| 7313| 0|
>>> 4767| 15191
>>>
>>> Is this a bug or feature please? Can someone throw the light?
>>> Good would be to print the units to the numbers too (ns).
>>>
>>
>>
>> That was likely a layout question of the latency tool's output. We could
>> simply dump something like "All latencies in nanoseconds" during
>> start-up. Would this be more helpful?
>>
>> Jan
>>
>>
>>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> Xenomai-help mailing list
>> Xenomai-help@domain.hid
>> https://mail.gna.org/listinfo/xenomai-help
>
>
>
--
Philippe.
^ permalink raw reply [flat|nested] 6+ messages in thread
* [Xenomai-core] Re: [Xenomai-help] negative values of latency/klatency
2006-02-28 12:33 ` Jan Kiszka
2006-02-28 12:52 ` Philippe Gerum
@ 2006-02-28 14:13 ` Jan Kiszka
2006-02-28 15:22 ` Philippe Gerum
1 sibling, 1 reply; 6+ messages in thread
From: Jan Kiszka @ 2006-02-28 14:13 UTC (permalink / raw)
To: xenomai-core; +Cc: xenomai, Rudolf Marek
[-- Attachment #1: Type: text/plain, Size: 2264 bytes --]
Jan Kiszka wrote:
> Rudolf Marek wrote:
>> ...
>> RTT| 00:00:01
>> RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>> RTD| 4767| 4929| 15191| 0| 4767| 15191
>> RTD| 4767| 4808| 8282| 0| 4767| 15191
>> RTD| 4767| 4808| 8080| 0| 4767| 15191
>> RTD| 4808| 4808| 7838| 0| 4767| 15191
>> RTD| 4767| 4808| 7272| 0| 4767| 15191
>> RTT| 00:00:06
>> RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>> RTD| 4808| 4808| 7555| 0| 4767| 15191
>> RTD| 4767| 4808| 7959| 0| 4767| 15191
>> RTD| 4767| 4808| 7393| 0| 4767| 15191
>> RTD| 4808| 4808| 7191| 0| 4767| 15191
>> RTD| 4767| 4808| 7313| 0| 4767| 15191
>>
>> Is this a bug or feature please? Can someone throw the light?
>> Good would be to print the units to the numbers too (ns).
>>
>
> That was likely a layout question of the latency tool's output. We could
> simply dump something like "All latencies in nanoseconds" during
> start-up. Would this be more helpful?
>
The output of latency is indeed inconsistent. Histogram and stats are
printed in microseconds, intermediate and overall latencies go out as
nanoseconds. Anyone any objections to switch to micros with 3 digits
after the decimal point? Patch is ready to be applied.
== Sampling period: 150 us
== Test mode: in-kernel timer handler
== All results in microseconds
warming up...
RTT| 00:00:01 (in-kernel timer handler, 150 us period)
RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat
worst
RTD| 8.935| 13.824| 39.951| 0| 8.935|
39.951
RTD| 8.998| 14.619| 36.867| 0| 8.935|
39.951
RTD| 8.576| 14.604| 37.417| 0| 8.576|
39.951
RTD| 3.018| 14.623| 40.466| 0| 3.018|
40.466
[grabbed on a low-end board]
Jan
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]
^ permalink raw reply [flat|nested] 6+ messages in thread
* [Xenomai-core] Re: [Xenomai-help] negative values of latency/klatency
2006-02-28 14:13 ` [Xenomai-core] " Jan Kiszka
@ 2006-02-28 15:22 ` Philippe Gerum
0 siblings, 0 replies; 6+ messages in thread
From: Philippe Gerum @ 2006-02-28 15:22 UTC (permalink / raw)
To: Jan Kiszka; +Cc: xenomai, Rudolf Marek, xenomai-core
Jan Kiszka wrote:
> Jan Kiszka wrote:
>
>>Rudolf Marek wrote:
>>
>>>...
>>>RTT| 00:00:01
>>>RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>>>RTD| 4767| 4929| 15191| 0| 4767| 15191
>>>RTD| 4767| 4808| 8282| 0| 4767| 15191
>>>RTD| 4767| 4808| 8080| 0| 4767| 15191
>>>RTD| 4808| 4808| 7838| 0| 4767| 15191
>>>RTD| 4767| 4808| 7272| 0| 4767| 15191
>>>RTT| 00:00:06
>>>RTH|----klat min|----klat avg|----klat max| overrun|---klat best|--klat worst
>>>RTD| 4808| 4808| 7555| 0| 4767| 15191
>>>RTD| 4767| 4808| 7959| 0| 4767| 15191
>>>RTD| 4767| 4808| 7393| 0| 4767| 15191
>>>RTD| 4808| 4808| 7191| 0| 4767| 15191
>>>RTD| 4767| 4808| 7313| 0| 4767| 15191
>>>
>>>Is this a bug or feature please? Can someone throw the light?
>>>Good would be to print the units to the numbers too (ns).
>>>
>>
>>That was likely a layout question of the latency tool's output. We could
>>simply dump something like "All latencies in nanoseconds" during
>>start-up. Would this be more helpful?
>>
>
>
> The output of latency is indeed inconsistent. Histogram and stats are
> printed in microseconds, intermediate and overall latencies go out as
> nanoseconds. Anyone any objections to switch to micros with 3 digits
> after the decimal point? Patch is ready to be applied.
Let's roll.
>
> == Sampling period: 150 us
> == Test mode: in-kernel timer handler
> == All results in microseconds
> warming up...
> RTT| 00:00:01 (in-kernel timer handler, 150 us period)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat
> worst
> RTD| 8.935| 13.824| 39.951| 0| 8.935|
> 39.951
> RTD| 8.998| 14.619| 36.867| 0| 8.935|
> 39.951
> RTD| 8.576| 14.604| 37.417| 0| 8.576|
> 39.951
> RTD| 3.018| 14.623| 40.466| 0| 3.018|
> 40.466
> [grabbed on a low-end board]
>
> Jan
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help
--
Philippe.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2006-02-28 15:22 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2006-02-27 16:12 [Xenomai-help] negative values of latency/klatency Rudolf Marek
2006-02-28 12:33 ` Jan Kiszka
2006-02-28 12:52 ` Philippe Gerum
2006-02-28 13:02 ` Philippe Gerum
2006-02-28 14:13 ` [Xenomai-core] " Jan Kiszka
2006-02-28 15:22 ` Philippe Gerum
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.