Re: [Xenomai] CyclicTest query on Xenomai

[Jan Kiszka <jan.kiszka@web.de>]

On 2015-06-30 12:45, Madan Sondur wrote:
> On Tue, Jun 30, 2015 at 11:10 AM, Jan Kiszka <jan.kiszka@web.de> wrote:
> 
>> Please don't top-post.
>>
>> On 2015-06-30 05:34, Madan Sondur wrote:
>>> I stress tested the system for an hour using Pi-stress and pumping huge
>>> amount of data through usb3.0 and I perf at network.  What I have
>> observed
>>> is cyclictest from usr/xenomai/bin behaves differently, only the first
>>> thread completes then after all the other threads execute very slowly,
>> you
>>> can see the count value.
>>>
>>> But the cyclictest that was downloaded from net completes all the threads
>>> and exits.
>>>
>>> The results of sample tests for 5000 loops are as shown.
>>>
>>> Xenomai cyclictest T1 count is 120, where as cyclictest which is from net
>>> has T1 count of 4546.
>>>
>>> Why is this behaviour?
>>
>> Xenomai uses different interval spreading than rt-test, see your
>> outputs. There is probably a bug in the Xenomai version of cyclictest,
>> getting some order of magnitude wrong with the default settings. Should
>> be easy to fix.
>>
> 
> Can you tell me what are the default settings for fixing this ?

cyclictest -d <distance>

> 
>>
>> The fact that you see over 300 µs worst-case latencies with vanilla
>> cyclictest on a standard kernel and that you don't get this with the
>> same test program over a Xenomai kernel, indicates that your test method
>> was not correct, e.g. that load was affecting only the other cores.
>> Xenomai does not improve the latencies of normal applications.
>>
> 
> 
> How do you say the vanilla cyclictest should not perform better in xenomai
> kernel than in standard kernel ?,
>   vanilla cyclictest is using posix calls which xenomai supports (posix api
> interface) for hard real time.
> 
> All the cores are shot up to 100% with pi_stress. only the first core was
> around 90% for traffic handling in usb 3.0 and Ethernet.
> 
> i don't think Vanilla cyclictest is a normal application.

Did you build the vanilla version against Xenomai libs using the
xeno-config provided parameters? Then the situation could be different.

Jan

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: OpenPGP digital signature
URL: <http://xenomai.org/pipermail/xenomai/attachments/20150630/d9439dac/attachment.sig>