[Xenomai-help] Cyclictest

[Xenomai-help] Cyclictest

[Perrine Martignoni]

[-- Attachment #1: Type: text/plain, Size: 2308 bytes --]

Hello,

I launched the same application Cyclictest on a kernel Linux without Xenomai
and a kernel Linux with Xenomai and I noticed that the result is different.

Here is the result with a kernel Linux without xenomai :

> cyclictest -t 10 -c 1 -i 10500


T: 0 (  732) P:99 I:   10500 C:    2291 Min:      41 Act:    2281 Avg:    5075
Max:   10224

T: 1 (  733) P:98 I:   11000 C:    2187 Min:      66 Act:    1312 Avg:    5085
Max:   10236

T: 2 (  734) P:97 I:   11500 C:    2092 Min:      91 Act:     842 Avg:    5113
Max:   10311

T: 3 (  735) P:96 I:   12000 C:    2004 Min:     106 Act:    1384 Avg:    5134
Max:   10140

T: 4 (  736) P:95 I:   12500 C:    1924 Min:     116 Act:    9873 Avg:    5152
Max:   10306

T: 5 (  737) P:94 I:   13000 C:    1850 Min:     136 Act:     415 Avg:    5172
Max:   10376

T: 6 (  738) P:93 I:   13500 C:    1782 Min:     136 Act:    3873 Avg:    5180
Max:   10411

T: 7 (  751) P:92 I:   14000 C:    1722 Min:     156 Act:    3559 Avg:    5184
Max:   10190

T: 8 (  752) P:91 I:   14500 C:    1663 Min:     140 Act:    8454 Avg:    5216
Max:   10310

T: 9 (  753) P:90 I:   15000 C:    1608 Min:     184 Act:    2485 Avg:    5214
Max:   10490


And here is the result with a kernel Linux with xenomai :
> cyclictest -t 10 -c 1 -i 10000


T: 0 (  893) P:99 I:   10000 C:    1756 Min:   13407 Act:   13536 Avg:   13452
Max:   13693

T: 1 (  894) P:98 I:   10500 C:    1673 Min:    4955 Act:    7595 Avg:    9767
Max:   14727

T: 2 (  895) P:97 I:   11000 C:    1597 Min:    5461 Act:    7628 Avg:   10047
Max:   14793

T: 3 (  908) P:96 I:   11500 C:    1333 Min:    4987 Act:    5641 Avg:    9808
Max:   14779

T: 4 (  909) P:95 I:   12000 C:    1277 Min:    5478 Act:   11665 Avg:    9588
Max:   13864

T: 5 (  910) P:94 I:   12500 C:    1226 Min:    6016 Act:   11191 Avg:    9859
Max:   14135

T: 6 (  911) P:93 I:   13000 C:    1179 Min:    5526 Act:    9711 Avg:   10130
Max:   14901

T: 7 (  912) P:92 I:   13500 C:    1135 Min:    5061 Act:   14740 Avg:    9896
Max:   14852

T: 8 (  913) P:91 I:   14000 C:    1095 Min:    5549 Act:    7766 Avg:    9669
Max:   14198

T: 9 (  914) P:90 I:   14500 C:    1057 Min:    5107 Act:   11820 Avg:    9940
Max:   14984


We can see that with Xenomai, the results are less good.
Is there an explication to this?

Thanks

[-- Attachment #2: Type: text/html, Size: 11989 bytes --]

Re: [Xenomai-help] Cyclictest

[Daniel Schnell]

Try adding -p99 to use realtime.
 
 
________________________________

From: xenomai-help-bounces@domain.hid [mailto:xenomai-help-bounces@domain.hid] On Behalf Of Perrine Martignoni
Sent: 6. júní 2007 15:02
To: xenomai-help
Subject: [Xenomai-help] Cyclictest


Hello, 
 
I launched the same application Cyclictest on a kernel Linux without Xenomai and a kernel Linux with Xenomai and I noticed that the result is different.
 
Here is the result with a kernel Linux without xenomai : 
 
> cyclictest -t 10 -c 1 -i 10500
  

T: 0 (  732) P:99 I:   10500 C:    2291 Min:      41 Act:    2281 Avg:    5075 Max:   10224 

T: 1 (  733) P:98 I:   11000 C:    2187 Min:      66 Act:    1312 Avg:    5085 Max:   10236 

T: 2 (  734) P:97 I:   11500 C:    2092 Min:      91 Act:     842 Avg:    5113 Max:   10311 

T: 3 (  735) P:96 I:   12000 C:    2004 Min:     106 Act:    1384 Avg:    5134 Max:   10140 

T: 4 (  736) P:95 I:   12500 C:    1924 Min:     116 Act:    9873 Avg:    5152 Max:   10306 

T: 5 (  737) P:94 I:   13000 C:    1850 Min:     136 Act:     415 Avg:    5172 Max:   10376 

T: 6 (  738) P:93 I:   13500 C:    1782 Min:     136 Act:    3873 Avg:    5180 Max:   10411 

T: 7 (  751) P:92 I:   14000 C:    1722 Min:     156 Act:    3559 Avg:    5184 Max:   10190 

T: 8 (  752) P:91 I:   14500 C:    1663 Min:     140 Act:    8454 Avg:    5216 Max:   10310 

T: 9 (  753) P:90 I:   15000 C:    1608 Min:     184 Act:    2485 Avg:    5214 Max:   10490 

 

And here is the result with a kernel Linux with xenomai : 
> cyclictest -t 10 -c 1 -i 10000
 

T: 0 (  893) P:99 I:   10000 C:     1756 Min:   13407 Act:   13536 Avg:   13452 Max:   13693 

T: 1 (  894) P:98 I:   10500 C:     1673 Min:    4955 Act:    7595 Avg:    9767 Max:   14727 

T: 2 (  895) P:97 I:   11000 C:     1597 Min:    5461 Act:    7628 Avg:   10047 Max:   14793 

T: 3 (  908) P:96 I:   11500 C:     1333 Min:    4987 Act:    5641 Avg:    9808 Max:   14779 

T: 4 (  909) P:95 I:   12000 C:     1277 Min:    5478 Act:   11665 Avg:    9588 Max:   13864 

T: 5 (  910) P:94 I:   12500 C:     1226 Min:    6016 Act:   11191 Avg:    9859 Max:   14135 

T: 6 (  911) P:93 I:   13000 C:     1179 Min:    5526 Act:    9711 Avg:   10130 Max:   14901 

T: 7 (  912) P:92 I:   13500 C:     1135 Min:    5061 Act:   14740 Avg:    9896 Max:   14852 

T: 8 (  913) P:91 I:   14000 C:     1095 Min:    5549 Act:    7766 Avg:    9669 Max:   14198 

T: 9 (  914) P:90 I:   14500 C:     1057 Min:    5107 Act:   11820 Avg:    9940 Max:   14984 

 
 
We can see that with Xenomai, the results are less good. 
Is there an explication to this?
 
Thanks

Re: [Xenomai-help] Cyclictest

[Philippe Gerum]

On Wed, 2007-06-06 at 15:04 +0000, Daniel Schnell wrote:
> Try adding -p99 to use realtime.
>  
> 
> ________________________________
> 
> From: xenomai-help-bounces@domain.hid [mailto:xenomai-help-bounces@domain.hid] On Behalf Of Perrine Martignoni
> Sent: 6. júní 2007 15:02
> To: xenomai-help
> Subject: [Xenomai-help] Cyclictest
> 
> 
> Hello, 
>  
> I launched the same application Cyclictest on a kernel Linux without Xenomai and a kernel Linux with Xenomai and I noticed that the result is different.
>  
> Here is the result with a kernel Linux without xenomai : 
>  
> > cyclictest -t 10 -c 1 -i 10500
>   
> 
> T: 0 (  732) P:99 I:   10500 C:    2291 Min:      41 Act:    2281 Avg:    5075 Max:   10224 
> 
> T: 1 (  733) P:98 I:   11000 C:    2187 Min:      66 Act:    1312 Avg:    5085 Max:   10236 
> 
> T: 2 (  734) P:97 I:   11500 C:    2092 Min:      91 Act:     842 Avg:    5113 Max:   10311 
> 
> T: 3 (  735) P:96 I:   12000 C:    2004 Min:     106 Act:    1384 Avg:    5134 Max:   10140 
> 
> T: 4 (  736) P:95 I:   12500 C:    1924 Min:     116 Act:    9873 Avg:    5152 Max:   10306 
> 
> T: 5 (  737) P:94 I:   13000 C:    1850 Min:     136 Act:     415 Avg:    5172 Max:   10376 
> 
> T: 6 (  738) P:93 I:   13500 C:    1782 Min:     136 Act:    3873 Avg:    5180 Max:   10411 
> 
> T: 7 (  751) P:92 I:   14000 C:    1722 Min:     156 Act:    3559 Avg:    5184 Max:   10190 
> 
> T: 8 (  752) P:91 I:   14500 C:    1663 Min:     140 Act:    8454 Avg:    5216 Max:   10310 
> 
> T: 9 (  753) P:90 I:   15000 C:    1608 Min:     184 Act:    2485 Avg:    5214 Max:   10490 
> 
>  
> 
> And here is the result with a kernel Linux with xenomai : 
> > cyclictest -t 10 -c 1 -i 10000
>  
> 
> T: 0 (  893) P:99 I:   10000 C:     1756 Min:   13407 Act:   13536 Avg:   13452 Max:   13693 
> 
> T: 1 (  894) P:98 I:   10500 C:     1673 Min:    4955 Act:    7595 Avg:    9767 Max:   14727 
> 
> T: 2 (  895) P:97 I:   11000 C:     1597 Min:    5461 Act:    7628 Avg:   10047 Max:   14793 
> 
> T: 3 (  908) P:96 I:   11500 C:     1333 Min:    4987 Act:    5641 Avg:    9808 Max:   14779 
> 
> T: 4 (  909) P:95 I:   12000 C:     1277 Min:    5478 Act:   11665 Avg:    9588 Max:   13864 
> 
> T: 5 (  910) P:94 I:   12500 C:     1226 Min:    6016 Act:   11191 Avg:    9859 Max:   14135 
> 
> T: 6 (  911) P:93 I:   13000 C:     1179 Min:    5526 Act:    9711 Avg:   10130 Max:   14901 
> 
> T: 7 (  912) P:92 I:   13500 C:     1135 Min:    5061 Act:   14740 Avg:    9896 Max:   14852 
> 
> T: 8 (  913) P:91 I:   14000 C:     1095 Min:    5549 Act:    7766 Avg:    9669 Max:   14198 
> 
> T: 9 (  914) P:90 I:   14500 C:     1057 Min:    5107 Act:   11820 Avg:    9940 Max:   14984 
> 
>  
> 
> We can see that with Xenomai, the results are less good. 
> Is there an explication to this?

Actually, you don't have any information from this test output unless
you did put some load on your box. Real-time is about predictability,
what Xenomai brings is exactly that. Raw performances have no meaning in
a real-time context unless they can be sustained regardless of the
runtime conditions.

If your box is not some flashy big iron, try putting some load with this
command while the test is running, on both boxen:

while :; do dd if=/dev/zero of=hog.out count=500 bs=1M; rm -f hog.out &&
sync; done

If you do have a recent, multi-core and multi-Ghz box, try compiling a
kernel over all the CPUs when testing. E.g. for a 2 CPU box, try
running:

while :; do make clean && make -j3; done

>  
> Thanks
> 
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help
-- 
Philippe.

Re: [Xenomai-help] Cyclictest

[Perrine Martignoni]

[-- Attachment #1: Type: text/plain, Size: 4383 bytes --]

>Did you double-check that you were not testing Xenomai with the
>"vanilla" cyclictest, ie. the one without libpthread_rt dependencies...?
>See "ldd cyclictest" when in doubt.

Yes, I did. I wanted to know performances of cyclictest without Xenomai to
compare.
And I thought that even with Xenomai in the kernel, I would have the same
result.

On 6/6/07, Philippe Gerum <rpm@xenomai.org> wrote:
>
> On Wed, 2007-06-06 at 15:04 +0000, Daniel Schnell wrote:
> > Try adding -p99 to use realtime.
> >
> >
> > ________________________________
> >
> > From: xenomai-help-bounces@domain.hid [mailto:xenomai-help-bounces@domain.hid]
> On Behalf Of Perrine Martignoni
> > Sent: 6. júní 2007 15:02
> > To: xenomai-help
> > Subject: [Xenomai-help] Cyclictest
> >
> >
> > Hello,
> >
> > I launched the same application Cyclictest on a kernel Linux without
> Xenomai and a kernel Linux with Xenomai and I noticed that the result is
> different.
> >
> > Here is the result with a kernel Linux without xenomai :
> >
> > > cyclictest -t 10 -c 1 -i 10500
> >
> >
> > T: 0 (  732) P:99 I:   10500 C:    2291 Min:      41 Act:    2281
> Avg:    5075 Max:   10224
> >
> > T: 1 (  733) P:98 I:   11000 C:    2187 Min:      66 Act:    1312
> Avg:    5085 Max:   10236
> >
> > T: 2 (  734) P:97 I:   11500 C:    2092 Min:      91 Act:     842
> Avg:    5113 Max:   10311
> >
> > T: 3 (  735) P:96 I:   12000 C:    2004 Min:     106 Act:    1384
> Avg:    5134 Max:   10140
> >
> > T: 4 (  736) P:95 I:   12500 C:    1924 Min:     116 Act:    9873
> Avg:    5152 Max:   10306
> >
> > T: 5 (  737) P:94 I:   13000 C:    1850 Min:     136 Act:     415
> Avg:    5172 Max:   10376
> >
> > T: 6 (  738) P:93 I:   13500 C:    1782 Min:     136 Act:    3873
> Avg:    5180 Max:   10411
> >
> > T: 7 (  751) P:92 I:   14000 C:    1722 Min:     156 Act:    3559
> Avg:    5184 Max:   10190
> >
> > T: 8 (  752) P:91 I:   14500 C:    1663 Min:     140 Act:    8454
> Avg:    5216 Max:   10310
> >
> > T: 9 (  753) P:90 I:   15000 C:    1608 Min:     184 Act:    2485
> Avg:    5214 Max:   10490
> >
> >
> >
> > And here is the result with a kernel Linux with xenomai :
> > > cyclictest -t 10 -c 1 -i 10000
> >
> >
> > T: 0 (  893) P:99 I:   10000 C:     1756 Min:   13407 Act:   13536
> Avg:   13452 Max:   13693
> >
> > T: 1 (  894) P:98 I:   10500 C:     1673 Min:    4955 Act:    7595
> Avg:    9767 Max:   14727
> >
> > T: 2 (  895) P:97 I:   11000 C:     1597 Min:    5461 Act:    7628
> Avg:   10047 Max:   14793
> >
> > T: 3 (  908) P:96 I:   11500 C:     1333 Min:    4987 Act:    5641
> Avg:    9808 Max:   14779
> >
> > T: 4 (  909) P:95 I:   12000 C:     1277 Min:    5478 Act:   11665
> Avg:    9588 Max:   13864
> >
> > T: 5 (  910) P:94 I:   12500 C:     1226 Min:    6016 Act:   11191
> Avg:    9859 Max:   14135
> >
> > T: 6 (  911) P:93 I:   13000 C:     1179 Min:    5526 Act:    9711
> Avg:   10130 Max:   14901
> >
> > T: 7 (  912) P:92 I:   13500 C:     1135 Min:    5061 Act:   14740
> Avg:    9896 Max:   14852
> >
> > T: 8 (  913) P:91 I:   14000 C:     1095 Min:    5549 Act:    7766
> Avg:    9669 Max:   14198
> >
> > T: 9 (  914) P:90 I:   14500 C:     1057 Min:    5107 Act:   11820
> Avg:    9940 Max:   14984
> >
> >
> >
> > We can see that with Xenomai, the results are less good.
> > Is there an explication to this?
>
> Actually, you don't have any information from this test output unless
> you did put some load on your box. Real-time is about predictability,
> what Xenomai brings is exactly that. Raw performances have no meaning in
> a real-time context unless they can be sustained regardless of the
> runtime conditions.
>
> If your box is not some flashy big iron, try putting some load with this
> command while the test is running, on both boxen:
>
> while :; do dd if=/dev/zero of=hog.out count=500 bs=1M; rm -f hog.out &&
> sync; done
>
> If you do have a recent, multi-core and multi-Ghz box, try compiling a
> kernel over all the CPUs when testing. E.g. for a 2 CPU box, try
> running:
>
> while :; do make clean && make -j3; done
>
> >
> > Thanks
> >
> > _______________________________________________
> > Xenomai-help mailing list
> > Xenomai-help@domain.hid
> > https://mail.gna.org/listinfo/xenomai-help
> --
> Philippe.
>
>
>

[-- Attachment #2: Type: text/html, Size: 7279 bytes --]

Re: [Xenomai-help] Cyclictest

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 525 bytes --]

Perrine Martignoni wrote:
>> Did you double-check that you were not testing Xenomai with the
>> "vanilla" cyclictest, ie. the one without libpthread_rt dependencies...?
>> See "ldd cyclictest" when in doubt.
> 
> Yes, I did. I wanted to know performances of cyclictest without Xenomai to
> compare.
> And I thought that even with Xenomai in the kernel, I would have the same
> result.

Then re-read Philippe's comment please.

If you want to test average performance, rather use something like lmbench.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

Re: [Xenomai-help] Cyclictest

[Perrine Martignoni]

[-- Attachment #1: Type: text/plain, Size: 770 bytes --]

Ok. I'll do this.

But I don't understand why the same application compiled without any links
with Xenomai give different results if there is Xenomai in the kernel.


On 6/7/07, Jan Kiszka <jan.kiszka@domain.hid> wrote:
>
> Perrine Martignoni wrote:
> >> Did you double-check that you were not testing Xenomai with the
> >> "vanilla" cyclictest, ie. the one without libpthread_rt
> dependencies...?
> >> See "ldd cyclictest" when in doubt.
> >
> > Yes, I did. I wanted to know performances of cyclictest without Xenomai
> to
> > compare.
> > And I thought that even with Xenomai in the kernel, I would have the
> same
> > result.
>
> Then re-read Philippe's comment please.
>
> If you want to test average performance, rather use something like
> lmbench.
>
> Jan
>
>
>

[-- Attachment #2: Type: text/html, Size: 1155 bytes --]

Re: [Xenomai-help] Cyclictest

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 490 bytes --]

Perrine Martignoni wrote:
> Ok. I'll do this.
> 
> But I don't understand why the same application compiled without any links
> with Xenomai give different results if there is Xenomai in the kernel.

[Looking at your numbers again] Hmm, maybe some rounding issue of ticks
due to whatever side-effect of I-pipe. We would first of all need the
usual set of information (.config, involved versions) and also
/sys/devices/system/clocksource/clocksource0/current_clocksource.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

Re: [Xenomai-help] Cyclictest

[Gilles Chanteperdrix]

Jan Kiszka wrote:
> Perrine Martignoni wrote:
> 
>>Ok. I'll do this.
>>
>>But I don't understand why the same application compiled without any links
>>with Xenomai give different results if there is Xenomai in the kernel.
> 
> 
> [Looking at your numbers again] Hmm, maybe some rounding issue of ticks
> due to whatever side-effect of I-pipe. We would first of all need the
> usual set of information (.config, involved versions) and also
> /sys/devices/system/clocksource/clocksource0/current_clocksource.

Maybe what Perrine is observing is simply the overhead of the I-pipe ?
I mean, Linux is Xenomai idle task, so it is acceptable for Linux
numbers to be a bit worse than when Xenomai is not running.

-- 
                                                 Gilles Chanteperdrix

Re: [Xenomai-help] Cyclictest

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 987 bytes --]

Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
>> Perrine Martignoni wrote:
>>
>>> Ok. I'll do this.
>>>
>>> But I don't understand why the same application compiled without any links
>>> with Xenomai give different results if there is Xenomai in the kernel.
>>
>> [Looking at your numbers again] Hmm, maybe some rounding issue of ticks
>> due to whatever side-effect of I-pipe. We would first of all need the
>> usual set of information (.config, involved versions) and also
>> /sys/devices/system/clocksource/clocksource0/current_clocksource.
> 
> Maybe what Perrine is observing is simply the overhead of the I-pipe ?
> I mean, Linux is Xenomai idle task, so it is acceptable for Linux
> numbers to be a bit worse than when Xenomai is not running.

I bit worse is expected. But I think we are seeing 1 or 2 ticks wake-up
delays here. As far as I understood, they are not due to Xenomai
consuming significant cycles in the background, are they, Perrine?

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

Re: [Xenomai-help] Cyclictest

[Gilles Chanteperdrix]

Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
> 
>>Jan Kiszka wrote:
>>
>>>Perrine Martignoni wrote:
>>>
>>>
>>>>Ok. I'll do this.
>>>>
>>>>But I don't understand why the same application compiled without any links
>>>>with Xenomai give different results if there is Xenomai in the kernel.
>>>
>>>[Looking at your numbers again] Hmm, maybe some rounding issue of ticks
>>>due to whatever side-effect of I-pipe. We would first of all need the
>>>usual set of information (.config, involved versions) and also
>>>/sys/devices/system/clocksource/clocksource0/current_clocksource.
>>
>>Maybe what Perrine is observing is simply the overhead of the I-pipe ?
>>I mean, Linux is Xenomai idle task, so it is acceptable for Linux
>>numbers to be a bit worse than when Xenomai is not running.
> 
> 
> I bit worse is expected. But I think we are seeing 1 or 2 ticks wake-up
> delays here. As far as I understood, they are not due to Xenomai
> consuming significant cycles in the background, are they, Perrine?

What I see is latencies 5us longer. 5us on ARM is almost nothing, keep
in mind that a simple syscall on ARM already takes 10 us.

-- 
                                                 Gilles Chanteperdrix

Re: [Xenomai-help] Cyclictest

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1670 bytes --]

Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
>> Gilles Chanteperdrix wrote:
>>
>>> Jan Kiszka wrote:
>>>
>>>> Perrine Martignoni wrote:
>>>>
>>>>
>>>>> Ok. I'll do this.
>>>>>
>>>>> But I don't understand why the same application compiled without any links
>>>>> with Xenomai give different results if there is Xenomai in the kernel.
>>>> [Looking at your numbers again] Hmm, maybe some rounding issue of ticks
>>>> due to whatever side-effect of I-pipe. We would first of all need the
>>>> usual set of information (.config, involved versions) and also
>>>> /sys/devices/system/clocksource/clocksource0/current_clocksource.
>>> Maybe what Perrine is observing is simply the overhead of the I-pipe ?
>>> I mean, Linux is Xenomai idle task, so it is acceptable for Linux
>>> numbers to be a bit worse than when Xenomai is not running.
>>
>> I bit worse is expected. But I think we are seeing 1 or 2 ticks wake-up
>> delays here. As far as I understood, they are not due to Xenomai
>> consuming significant cycles in the background, are they, Perrine?
> 
> What I see is latencies 5us longer. 5us on ARM is almost nothing, keep
> in mind that a simple syscall on ARM already takes 10 us.
> 

T: 0 (  732) P:99 I:   10500 C:    2291 Min:      41 Act:    2281 Avg:    5075 Max:   10224 

vs.

T: 0 (  893) P:99 I:   10000 C:    1756 Min:   13407 Act:   13536 Avg:   13452 Max:   13693 

Thus the _minimum_ and the _average_ wakeup time of a Linux task
increase fairly significant here. That's the point. Given a reasonable
runtime of both test, this /could/ be considered as a regression of a
Xenomai-enabled Linux kernel.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

Re: [Xenomai-help] Cyclictest

[Gilles Chanteperdrix]

Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
> 
>>Jan Kiszka wrote:
>>
>>>Gilles Chanteperdrix wrote:
>>>
>>>
>>>>Jan Kiszka wrote:
>>>>
>>>>
>>>>>Perrine Martignoni wrote:
>>>>>
>>>>>
>>>>>
>>>>>>Ok. I'll do this.
>>>>>>
>>>>>>But I don't understand why the same application compiled without any links
>>>>>>with Xenomai give different results if there is Xenomai in the kernel.
>>>>>
>>>>>[Looking at your numbers again] Hmm, maybe some rounding issue of ticks
>>>>>due to whatever side-effect of I-pipe. We would first of all need the
>>>>>usual set of information (.config, involved versions) and also
>>>>>/sys/devices/system/clocksource/clocksource0/current_clocksource.
>>>>
>>>>Maybe what Perrine is observing is simply the overhead of the I-pipe ?
>>>>I mean, Linux is Xenomai idle task, so it is acceptable for Linux
>>>>numbers to be a bit worse than when Xenomai is not running.
>>>
>>>I bit worse is expected. But I think we are seeing 1 or 2 ticks wake-up
>>>delays here. As far as I understood, they are not due to Xenomai
>>>consuming significant cycles in the background, are they, Perrine?
>>
>>What I see is latencies 5us longer. 5us on ARM is almost nothing, keep
>>in mind that a simple syscall on ARM already takes 10 us.
>>
> 
> 
> T: 0 (  732) P:99 I:   10500 C:    2291 Min:      41 Act:    2281 Avg:    5075 Max:   10224 
> 
> vs.
> 
> T: 0 (  893) P:99 I:   10000 C:    1756 Min:   13407 Act:   13536 Avg:   13452 Max:   13693 
> 
> Thus the _minimum_ and the _average_ wakeup time of a Linux task
> increase fairly significant here. That's the point. Given a reasonable
> runtime of both test, this /could/ be considered as a regression of a
> Xenomai-enabled Linux kernel.

Ok, results are in milliseconds, not microseconds. I guess what we
observe here is an effect of Xenomai timer interception.

Perrine, what version of the I-pipe patch are you using ?

-- 
                                                 Gilles Chanteperdrix

Re: [Xenomai-help] Cyclictest

[Philippe Gerum]

On Thu, 2007-06-07 at 11:25 +0200, Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
> > Perrine Martignoni wrote:
> > 
> >>Ok. I'll do this.
> >>
> >>But I don't understand why the same application compiled without any links
> >>with Xenomai give different results if there is Xenomai in the kernel.
> > 
> > 
> > [Looking at your numbers again] Hmm, maybe some rounding issue of ticks
> > due to whatever side-effect of I-pipe. We would first of all need the
> > usual set of information (.config, involved versions) and also
> > /sys/devices/system/clocksource/clocksource0/current_clocksource.
> 
> Maybe what Perrine is observing is simply the overhead of the I-pipe ?
> I mean, Linux is Xenomai idle task, so it is acceptable for Linux
> numbers to be a bit worse than when Xenomai is not running.
> 

Most of the "known" overhead has disappeared with the PF_EVNOTIFY
extension, so that we relay Linux-originated events (e.g. syscalls) only
to subscribed domains (e.g. Xenomai). In the test case we are discussing
of, no relay should be done since no domain did subscribe to receive
those events.

The best way to check if the I-pipe still adds a noticeable overhead is
to compare:

- a vanilla kernel running cyclic test, 
with
- an I-pipe patched kernel, without Xenomai, using the same .config than
the vanilla kernel used. Without Xenomai either means to not run the
prepare-kernel.sh script, but only patch the kernel with the I-pipe
support, or disable CONFIG_XENOMAI entirely if the preparation script
has been run on the target kernel tree.

_Having both kernel setups share the very same kernel configuration is
mandatory, otherwise you would end up comparing apples and oranges._
The best way to make sure of this is to copy the .config file used for
the vanilla kernel to the I-pipe patched one, and run "make oldconfig".
You should be prompted only once, for the new CONFIG_IPIPE switch. Any
other behaviour would be the sign of some discrepancy in your test
configuration.

Since we can't have any temporal predictability for such configs, the
max value the cyclic test reports has no meaning. The useful figures are
the min and average values, as observed for a few minutes while the
board/box is sitting doing nothing but running the test.

A better way measure such artefacts is to run something like hackbench,
and compare the execution times, on kernels w/ and w/o the I-pipe.
http://developer.osdl.org/craiger/hackbench/src/hackbench.c
Tune the argument representing the number of process groups which you
pass to this program, so that it runs for a few minutes without putting
your box/board on its knees. There should be no significant difference
between the time reported at the end of both tests, at the very least,
not a 30% increase as your first report suggested. If the difference is
noticeable, then, there might be an issue (the final conclusion really
depends on the underlying hw).

-- 
Philippe.

Re: [Xenomai-help] Cyclictest

[Philippe Gerum]

On Thu, 2007-06-07 at 12:07 +0200, Philippe Gerum wrote:
> On Thu, 2007-06-07 at 11:25 +0200, Gilles Chanteperdrix wrote:
> > Jan Kiszka wrote:
> > > Perrine Martignoni wrote:
> > > 
> > >>Ok. I'll do this.
> > >>
> > >>But I don't understand why the same application compiled without any links
> > >>with Xenomai give different results if there is Xenomai in the kernel.
> > > 
> > > 
> > > [Looking at your numbers again] Hmm, maybe some rounding issue of ticks
> > > due to whatever side-effect of I-pipe. We would first of all need the
> > > usual set of information (.config, involved versions) and also
> > > /sys/devices/system/clocksource/clocksource0/current_clocksource.
> > 
> > Maybe what Perrine is observing is simply the overhead of the I-pipe ?
> > I mean, Linux is Xenomai idle task, so it is acceptable for Linux
> > numbers to be a bit worse than when Xenomai is not running.
> > 
> 
> Most of the "known" overhead has disappeared with the PF_EVNOTIFY
> extension, so that we relay Linux-originated events (e.g. syscalls) only
> to subscribed domains (e.g. Xenomai).

... to subscribed tasks from suscribed domains, more precisely. In
effect, Xenomai does not filter syscalls issued from plain Linux tasks
anymore.

>  In the test case we are discussing
> of, no relay should be done since no domain did subscribe to receive
> those events.
> 
> The best way to check if the I-pipe still adds a noticeable overhead is
> to compare:
> 
> - a vanilla kernel running cyclic test, 
> with
> - an I-pipe patched kernel, without Xenomai, using the same .config than
> the vanilla kernel used. Without Xenomai either means to not run the
> prepare-kernel.sh script, but only patch the kernel with the I-pipe
> support, or disable CONFIG_XENOMAI entirely if the preparation script
> has been run on the target kernel tree.
> 
> _Having both kernel setups share the very same kernel configuration is
> mandatory, otherwise you would end up comparing apples and oranges._
> The best way to make sure of this is to copy the .config file used for
> the vanilla kernel to the I-pipe patched one, and run "make oldconfig".
> You should be prompted only once, for the new CONFIG_IPIPE switch. Any
> other behaviour would be the sign of some discrepancy in your test
> configuration.
> 
> Since we can't have any temporal predictability for such configs, the
> max value the cyclic test reports has no meaning. The useful figures are
> the min and average values, as observed for a few minutes while the
> board/box is sitting doing nothing but running the test.
> 
> A better way measure such artefacts is to run something like hackbench,
> and compare the execution times, on kernels w/ and w/o the I-pipe.
> http://developer.osdl.org/craiger/hackbench/src/hackbench.c
> Tune the argument representing the number of process groups which you
> pass to this program, so that it runs for a few minutes without putting
> your box/board on its knees. There should be no significant difference
> between the time reported at the end of both tests, at the very least,
> not a 30% increase as your first report suggested. If the difference is
> noticeable, then, there might be an issue (the final conclusion really
> depends on the underlying hw).
> 
-- 
Philippe.

Re: [Xenomai-help] Cyclictest

[Perrine Martignoni]

[-- Attachment #1: Type: text/plain, Size: 3711 bytes --]

> Perrine, what version of the I-pipe patch are you using ?

I use adeos-ipipe-2.6.19-arm-1.6-04.patch.


On 6/7/07, Philippe Gerum <rpm@xenomai.org> wrote:
>
> On Thu, 2007-06-07 at 12:07 +0200, Philippe Gerum wrote:
> > On Thu, 2007-06-07 at 11:25 +0200, Gilles Chanteperdrix wrote:
> > > Jan Kiszka wrote:
> > > > Perrine Martignoni wrote:
> > > >
> > > >>Ok. I'll do this.
> > > >>
> > > >>But I don't understand why the same application compiled without any
> links
> > > >>with Xenomai give different results if there is Xenomai in the
> kernel.
> > > >
> > > >
> > > > [Looking at your numbers again] Hmm, maybe some rounding issue of
> ticks
> > > > due to whatever side-effect of I-pipe. We would first of all need
> the
> > > > usual set of information (.config, involved versions) and also
> > > > /sys/devices/system/clocksource/clocksource0/current_clocksource.
> > >
> > > Maybe what Perrine is observing is simply the overhead of the I-pipe ?
> > > I mean, Linux is Xenomai idle task, so it is acceptable for Linux
> > > numbers to be a bit worse than when Xenomai is not running.
> > >
> >
> > Most of the "known" overhead has disappeared with the PF_EVNOTIFY
> > extension, so that we relay Linux-originated events (e.g. syscalls) only
> > to subscribed domains (e.g. Xenomai).
>
> ... to subscribed tasks from suscribed domains, more precisely. In
> effect, Xenomai does not filter syscalls issued from plain Linux tasks
> anymore.
>
> >  In the test case we are discussing
> > of, no relay should be done since no domain did subscribe to receive
> > those events.
> >
> > The best way to check if the I-pipe still adds a noticeable overhead is
> > to compare:
> >
> > - a vanilla kernel running cyclic test,
> > with
> > - an I-pipe patched kernel, without Xenomai, using the same .config than
> > the vanilla kernel used. Without Xenomai either means to not run the
> > prepare-kernel.sh script, but only patch the kernel with the I-pipe
> > support, or disable CONFIG_XENOMAI entirely if the preparation script
> > has been run on the target kernel tree.
> >
> > _Having both kernel setups share the very same kernel configuration is
> > mandatory, otherwise you would end up comparing apples and oranges._
> > The best way to make sure of this is to copy the .config file used for
> > the vanilla kernel to the I-pipe patched one, and run "make oldconfig".
> > You should be prompted only once, for the new CONFIG_IPIPE switch. Any
> > other behaviour would be the sign of some discrepancy in your test
> > configuration.
> >
> > Since we can't have any temporal predictability for such configs, the
> > max value the cyclic test reports has no meaning. The useful figures are
> > the min and average values, as observed for a few minutes while the
> > board/box is sitting doing nothing but running the test.
> >
> > A better way measure such artefacts is to run something like hackbench,
> > and compare the execution times, on kernels w/ and w/o the I-pipe.
> > http://developer.osdl.org/craiger/hackbench/src/hackbench.c
> > Tune the argument representing the number of process groups which you
> > pass to this program, so that it runs for a few minutes without putting
> > your box/board on its knees. There should be no significant difference
> > between the time reported at the end of both tests, at the very least,
> > not a 30% increase as your first report suggested. If the difference is
> > noticeable, then, there might be an issue (the final conclusion really
> > depends on the underlying hw).
> >
> --
> Philippe.
>
>
>
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help
>

[-- Attachment #2: Type: text/html, Size: 4609 bytes --]

Re: [Xenomai-help] Cyclictest

[Gilles Chanteperdrix]

Perrine Martignoni wrote:
>> Perrine, what version of the I-pipe patch are you using ?
>  
> I use adeos-ipipe-2.6.19-arm-1.6-04.patch.

Please switch to the 1.7 serie and tell us if you observe the same problems.

-- 
                                                 Gilles Chanteperdrix

Re: [Xenomai-help] Cyclictest

[Perrine Martignoni]

[-- Attachment #1: Type: text/plain, Size: 397 bytes --]

Ok, I took it.

On 6/7/07, Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org> wrote:
>
> Perrine Martignoni wrote:
> >> Perrine, what version of the I-pipe patch are you using ?
> >
> > I use adeos-ipipe-2.6.19-arm-1.6-04.patch.
>
> Please switch to the 1.7 serie and tell us if you observe the same
> problems.
>
> --
>                                                 Gilles Chanteperdrix
>

[-- Attachment #2: Type: text/html, Size: 931 bytes --]

Re: [Xenomai-help] Cyclictest

[Gilles Chanteperdrix]

Perrine Martignoni wrote:
> Ok, I took it.

Beware, you also need to pick the latest version of Xenomai from the
repository (trunk or v2.3.x branch).

-- 
                                                 Gilles Chanteperdrix

Re: [Xenomai-help] Cyclictest

[Perrine Martignoni]

[-- Attachment #1: Type: text/plain, Size: 411 bytes --]

With adeos 1.7, I have the same result.
I have Xenomai 2.3.1, is there a more recent version?


On 6/7/07, Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org> wrote:
>
> Perrine Martignoni wrote:
> > Ok, I took it.
>
> Beware, you also need to pick the latest version of Xenomai from the
> repository (trunk or v2.3.x branch).
>
> --
>                                                 Gilles Chanteperdrix
>

[-- Attachment #2: Type: text/html, Size: 971 bytes --]

Re: [Xenomai-help] Cyclictest

[Gilles Chanteperdrix]

Perrine Martignoni wrote:
> With adeos 1.7, I have the same result.
> I have Xenomai 2.3.1, is there a more recent version?

Do not do that ! You should not use Xenomai 2.3.1 with Adeos 1.7. Pick
Xenomai from the svn repository or stop using adeos 1.7.

-- 
                                                 Gilles Chanteperdrix

Re: [Xenomai-help] Cyclictest

[Gilles Chanteperdrix]

Gilles Chanteperdrix wrote:
> Perrine Martignoni wrote:
> 
>>With adeos 1.7, I have the same result.
>>I have Xenomai 2.3.1, is there a more recent version?
> 
> 
> Do not do that ! You should not use Xenomai 2.3.1 with Adeos 1.7. Pick
> Xenomai from the svn repository or stop using adeos 1.7.

I mean adeos 1.7 for ARM. Do not use Adeos 1.7 *for ARM* with Xenomai 2.3.1.

-- 
                                                 Gilles Chanteperdrix

Re: [Xenomai-help] Cyclictest

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 2809 bytes --]

Perrine Martignoni wrote:
> Hello,
> 
> I launched the same application Cyclictest on a kernel Linux without
> Xenomai
> and a kernel Linux with Xenomai and I noticed that the result is different.
> 
> Here is the result with a kernel Linux without xenomai :
> 
>> cyclictest -t 10 -c 1 -i 10500
> 
> 
> T: 0 (  732) P:99 I:   10500 C:    2291 Min:      41 Act:    2281
> Avg:    5075
> Max:   10224
> 
> T: 1 (  733) P:98 I:   11000 C:    2187 Min:      66 Act:    1312
> Avg:    5085
> Max:   10236
> 
> T: 2 (  734) P:97 I:   11500 C:    2092 Min:      91 Act:     842
> Avg:    5113
> Max:   10311
> 
> T: 3 (  735) P:96 I:   12000 C:    2004 Min:     106 Act:    1384
> Avg:    5134
> Max:   10140
> 
> T: 4 (  736) P:95 I:   12500 C:    1924 Min:     116 Act:    9873
> Avg:    5152
> Max:   10306
> 
> T: 5 (  737) P:94 I:   13000 C:    1850 Min:     136 Act:     415
> Avg:    5172
> Max:   10376
> 
> T: 6 (  738) P:93 I:   13500 C:    1782 Min:     136 Act:    3873
> Avg:    5180
> Max:   10411
> 
> T: 7 (  751) P:92 I:   14000 C:    1722 Min:     156 Act:    3559
> Avg:    5184
> Max:   10190
> 
> T: 8 (  752) P:91 I:   14500 C:    1663 Min:     140 Act:    8454
> Avg:    5216
> Max:   10310
> 
> T: 9 (  753) P:90 I:   15000 C:    1608 Min:     184 Act:    2485
> Avg:    5214
> Max:   10490
> 
> 
> And here is the result with a kernel Linux with xenomai :
>> cyclictest -t 10 -c 1 -i 10000
> 
> 
> T: 0 (  893) P:99 I:   10000 C:    1756 Min:   13407 Act:   13536 Avg:  
> 13452
> Max:   13693
> 
> T: 1 (  894) P:98 I:   10500 C:    1673 Min:    4955 Act:    7595
> Avg:    9767
> Max:   14727
> 
> T: 2 (  895) P:97 I:   11000 C:    1597 Min:    5461 Act:    7628 Avg:  
> 10047
> Max:   14793
> 
> T: 3 (  908) P:96 I:   11500 C:    1333 Min:    4987 Act:    5641
> Avg:    9808
> Max:   14779
> 
> T: 4 (  909) P:95 I:   12000 C:    1277 Min:    5478 Act:   11665
> Avg:    9588
> Max:   13864
> 
> T: 5 (  910) P:94 I:   12500 C:    1226 Min:    6016 Act:   11191
> Avg:    9859
> Max:   14135
> 
> T: 6 (  911) P:93 I:   13000 C:    1179 Min:    5526 Act:    9711 Avg:  
> 10130
> Max:   14901
> 
> T: 7 (  912) P:92 I:   13500 C:    1135 Min:    5061 Act:   14740
> Avg:    9896
> Max:   14852
> 
> T: 8 (  913) P:91 I:   14000 C:    1095 Min:    5549 Act:    7766
> Avg:    9669
> Max:   14198
> 
> T: 9 (  914) P:90 I:   14500 C:    1057 Min:    5107 Act:   11820
> Avg:    9940
> Max:   14984
> 
> 
> We can see that with Xenomai, the results are less good.
> Is there an explication to this?

Did you double-check that you were not testing Xenomai with the
"vanilla" cyclictest, ie. the one without libpthread_rt dependencies...?
See "ldd cyclictest" when in doubt.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]