[Fedora-music-list] Low Latency vs. Real Time Kernel - actual latencies ?

[Brian Monroe]

Bah, I forgot to save before I uploaded, here's the text file with the
CCRMA results.

On Tue, Feb 21, 2012 at 10:20 PM, Brian Monroe <briancmonroe at gmail.com>wrote:

> I didn't have time tonight to get tests on all the kernels under load, but
> here are the results, relatively load free.
>
> I need to write a short python program to put the systems under duress.
> Hopefully I'll have time tommorrow to work on it and I can have the rest of
> the results in.
>
> PS:I'm attaching some results from an RT kernel on a debian system below
> that they used while the box was recieving flood ping from an external
> source and repeated loops of hackbench 25 and ls -Ral /.
>
> # cyclictest -a -t -n -p99 -i100 -d50
> 560.44 586.11 606.12 211/1160 3727
> T: 0 (18617) P:99 I:100 C:1011846111 Min:  2 Act:  4 Avg:  5 Max: 39
> T: 1 (18618) P:98 I:150 C: 708641019 Min:  2 Act:  5 Avg: 11 Max: 57
>
>
>
> On Thu, Feb 16, 2012 at 10:37 AM, Fernando Lopez-Lezcano <
> nando at ccrma.stanford.edu> wrote:
>
>> On 02/16/2012 04:33 AM, David Timms wrote:
>>
>>> On 15/02/12 16:49, Brian Monroe wrote:
>>>
>>>> On Tue, Feb 14, 2012 at 12:21 PM, Brendan Jones
>>>> <brendan.jones.it at gmail.com>**wrote:
>>>>
>>>>> Latency times are really relative things based on many factors. Have
>>>>> you
>>>>> been getting better latency times than ta
>>>>>
>>>>>  yeah.
>>>>
>>>
>>> I would be good to see real numbers associated with claims like the
>>> above; also how it is measured..
>>>
>>
>> Cyclictest is widely used for measuring latencies:
>>  https://rt.wiki.kernel.org/**articles/c/y/c/Cyclictest.html<https://rt.wiki.kernel.org/articles/c/y/c/Cyclictest.html>
>>
>>
>>  Also, indicate those various kernel patches and or options they use or
>>> have tested.
>>>
>>
>> The rt patches for the latest kernels (3.2 at this point) are here:
>>  http://www.kernel.org/pub/**linux/kernel/projects/rt/3.2/<http://www.kernel.org/pub/linux/kernel/projects/rt/3.2/>
>>
>> This is pretty much all I add to the rt patched kernels for Fedora (with
>> the proper configuration options, of course, I'm attaching the extra
>> options I'm currently using). To get the best performance in the Jack world
>> you need to tune the irq kernel thread priorities, that is usually done
>> using rtirq (http://www.rncbc.org/jack/#**rtirq<http://www.rncbc.org/jack/#rtirq>),
>> and of course jackd should run with the proper rt priority as well.
>>
>> -- Fernando
>>
>> _______________________________________________
>> music mailing list
>> music at lists.fedoraproject.org
>> https://admin.fedoraproject.org/mailman/listinfo/music
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.fedoraproject.org/pipermail/music/attachments/20120221/2afdafb9/attachment-0001.html>
-------------- next part --------------
Stock 3.2.3-1- no load. 
[root at crappylappy ~]# cyclictest -a -t -n -p99 -i100 -d50 -l1000000 -q
# /dev/cpu_dma_latency set to 0us
T: 0 ( 2007) P:99 I:100 C:1000000 Min:      4 Act:   16 Avg:   16 Max:    3339
T: 1 ( 2008) P:98 I:150 C: 666773 Min:      4 Act:   16 Avg:   19 Max:   16543

My 3.2.3-1 Low Latency Kernel - No Load.
[root at crappylappy ~]# cyclictest -a -t -n -p99 -i100 -d50 -l1000000 -q
# /dev/cpu_dma_latency set to 0us
T: 0 ( 1796) P:99 I:100 C:1000000 Min:      4 Act:   16 Avg:   16 Max:     612
T: 1 ( 1797) P:98 I:150 C: 666769 Min:      4 Act:   16 Avg:   16 Max:     274

CCRMA RT Kernel - No Load
[root at crappylappy ~]# cyclictest -a -t -n -p99 -i100 -d50 -l1000000 -q
# /dev/cpu_dma_latency set to 0us
T: 0 ( 1872) P:99 I:100 C:1000000 Min:      6 Act:   26 Avg:   23 Max:      98
T: 1 ( 1873) P:98 I:150 C: 666682 Min:      6 Act:   20 Avg:   23 Max:      74