[Freeswitch-users] KVM vs. XEN virtualization

Dmitry Lysenko dvl36.ripe.nick at gmail.com
Thu Apr 9 21:09:40 MSD 2015 

2015-04-08 15:25 GMT+03:00 Sergey Safarov <s.safarov at gmail.com>:

> 1) 125uS is the interval between FrameSync of ONE(!) PCM sample at 8000hz,
> But here we are talking about packet telephony where commonly used frame
> length is 20000 uS.
> So, intervals for testing should be 5000-10000-20000uS, not 125uS.
>
> Not all FS installation has SIP endpoint. Some can have TDM cards, USB
> dongle and other analog equipment. Can you say what is length of this frame
> type?
> I am wondering about this question. I know that the frame 125ms is enough.
>

Most(if not all) of modern equipment have buffers. TDM cards too. Even
$0.25 microcontrollers have buffers.
Firing interrupts at each PCM sample not needed.
If you want massive sync audio processing (w/o buffering) you can use hard
realtime OS. PC and Linux (even on bare metal) is not suitable for this, at
least without RT patched kernel.


> 2) Why you decided that all of 42 threads should wake up simultaneously?
> That totally impossible in real world applications.
> Here should be tens of uS. Dependent of network card and, possibly, other
> I/O hardware.
>
> Can you say how many FS create threads for 40-50 active calls?
>


AFAIR, 1 thread per call leg, but depends of endpoints. Most of time
threads in I/O waiting/sleeping and encoding/decoding state.


> 3) -q switch is missing
> I am thinking it is optional. Result is same
>

Dynamic output to screen consumes resources. How much, depends on
method.(VGA, Serial console, SSH). In my case difference is significiant.
So, fair testing should exclude such things, IMO.


> 4) Precise time measurement and realtime processing in VM environment  is
> a challenge.
> To obtain acceptable results fine tuning of host's and guest's  required.
> WIthout that this numbers ... does not mean anything.
>
> Can you learn me how to correctly measure timer source quality in VM?
>

Serious players, as Wind River, use external timer.

In your case, most capable realtime hypervisor with RT_PREEMPT guest would
be one of the best solution.

Dmitry.


> On Wed, Apr 8, 2015 at 2:07 PM, Dmitry Lysenko <dvl36.ripe.nick at gmail.com>
> wrote:
>
>> 2015-04-07 10:48 GMT+03:00 Sergey Safarov <s.safarov at gmail.com>:
>>
>>> ...
>>> 2) i 125 - time us between software timer interrupts for first thread.
>>> It time is equal time between samples rate in PSTN networks
>>> In order to maintain our sampling rate we have 125 microseconds (uS)
>>> between consecutive samples (1/8,000Hz), which is more
>>>
>> ...
>>>
>>
>> 125uS is the interval between FrameSync of ONE(!) PCM sample at 8000hz,
>> But here we are talking about packet telephony where commonly used frame
>> length is 20000 uS.
>> So, intervals for testing should be 5000-10000-20000uS, not 125uS.
>>
>>
>>> 3) d 0 - this switch will make time between software timer interrupts
>>> for second and following threads is equal timer value for first thread
>>>
>>
>> Why you decided that all of 42 threads should wake up simultaneously?
>> That totally impossible in real world applications.
>> Here should be tens of uS. Dependent of network card and, possibly, other
>> I/O hardware.
>>
>>
>>> 4) l 240000 - how many timer test must be made. It is adjusted for 30 s
>>> overall test
>>> 5) n - use clock_nanosleep. it is allow take more accurate timer test
>>> values on my VM
>>> 6) p 80 - it is copied from example on Cyclictest page
>>>
>>
>> -q switch is missing.
>>
>>
>>> And how in your opinion this numbers are related to Freeswitch in VM
>>> environment?
>>> Relation between Cyclictest switches and FS described above.
>>> Relation betwen Cyclictest results and FS simple
>>>
>>
>> Precise time measurement and realtime processing in VM environment  is a
>> challenge.
>> To obtain acceptable results fine tuning of host's and guest's  required.
>> WIthout that this numbers ... does not mean anything.
>>
>>
>> On Tue, Apr 7, 2015 at 5:38 AM, Dmitry Lysenko <dvl36.ripe.nick at gmail.com
>>> > wrote:
>>>
>>>> Sergey,
>>>> Why you suggest to use exactly this numbers in cyclictest's switches?
>>>> And how in your opinion this numbers are related to Freeswitch in VM
>>>> environment?
>>>>
>>>> P.S.
>>>> https://mindlinux.wordpress.com/2013/10/25/using-and-understanding-the-real-time-cyclictest-benchmark-frank-rowand-sony/
>>>>
>>>> 2015-04-06 15:27 GMT+03:00 Sergey Safarov <s.safarov at gmail.com>:
>>>>
>>>>> Errol can you test VM timer with utility
>>>>> https://rt.wiki.kernel.org/index.php/Cyclictest
>>>>>
>>>>> My VM currently have folowing results
>>>>>
>>>>> [root at fs1 ~]#  date && /tmp/rt-tests/cyclictest -t42 -p 80 -d 0 -n -i 125 -l 240000 && date
>>>>> Sun Apr  5 07:01:31 GMT 2015
>>>>> # /dev/cpu_dma_latency set to 0us
>>>>> policy: fifo: loadavg: 2.38 0.93 0.50 1/194 1574
>>>>>
>>>>> T: 0 ( 1531) P:80 I:125 C: 240000 Min:      2 Act:   12 Avg:   16 Max:    1302
>>>>> T: 1 ( 1532) P:80 I:125 C: 240000 Min:      2 Act:   15 Avg:   14 Max:    2895
>>>>> T: 2 ( 1533) P:80 I:125 C: 240000 Min:      3 Act:   10 Avg:   12 Max:    2888
>>>>> T: 3 ( 1534) P:80 I:125 C: 240000 Min:      1 Act:   16 Avg:   13 Max:    1333
>>>>> T: 4 ( 1535) P:80 I:125 C: 240000 Min:      2 Act:   20 Avg:   14 Max:    2962
>>>>> T: 5 ( 1536) P:80 I:125 C: 240000 Min:      1 Act:   12 Avg:   14 Max:    2946
>>>>> T: 6 ( 1537) P:80 I:125 C: 240000 Min:      2 Act:   10 Avg:   13 Max:    1363
>>>>> T: 7 ( 1538) P:80 I:125 C: 240000 Min:      2 Act:   12 Avg:   13 Max:    1344
>>>>> T: 8 ( 1539) P:80 I:125 C: 240000 Min:      3 Act:   12 Avg:   16 Max:    1408
>>>>> T: 9 ( 1540) P:80 I:125 C: 240000 Min:      1 Act:    8 Avg:   13 Max:    2882
>>>>> T:10 ( 1541) P:80 I:125 C: 239974 Min:      2 Act:   12 Avg:   13 Max:    2908
>>>>>
>>>>>
>>>>> On Mon, Apr 6, 2015 at 11:15 AM, Errol Samuels <ewsamuels at gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Good job and Thanks for sharing!
>>>>>>
>>>>>> KVM at digitalocean.com, Debian 7 64bit, FreeSWITCH 1.4.15: bad
>>>>>>> precision
>>>>>>
>>>>>>
>>>>>> As someone who has been using KVM for a few years and has just
>>>>>> migrated about 20 VMs across to Digital Ocean about 6 weeks ago I admit I
>>>>>> am a little concerned about this finding.
>>>>>>
>>>>>> Just out of curiosity can you tell me the specs of the VM that you
>>>>>> tested with and which DO Datacenter?  I am using NYC3 so perhaps and I will
>>>>>> do some tests as soon as I have some time.
>>>>>>
>>>>>> With regards to Xen was this at Linode or inhouse?
>>>>>>
>>>>>> Thanks
>>>>>>
>>>>>> Errol
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Sun, Apr 5, 2015 at 1:37 AM, Stanislav Sinyagin <
>>>>>> ssinyagin at gmail.com> wrote:
>>>>>>
>>>>>>> here we go:
>>>>>>>
>>>>>>> https://txlab.wordpress.com/2015/04/05/testing-rtp-clock-precision-on-a-virtual-machine/
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On Sat, Apr 4, 2015 at 4:26 PM, Stanislav Sinyagin <
>>>>>>> ssinyagin at gmail.com> wrote:
>>>>>>> > On Sat, Apr 4, 2015 at 1:03 PM, Regis M <
>>>>>>> regis.freeswitch.org at tornad.net> wrote:
>>>>>>> >> How do you check wavs together in audacity ?
>>>>>>> >
>>>>>>> > you just open two WAV files, align their starting position, and
>>>>>>> > visually check the waves. If there's a problem, you would see that
>>>>>>> > after few seconds, the waves are not aligned any more.
>>>>>>> >
>>>>>>> >> Can you send me 2 samples wav (send... and recorded) so I could
>>>>>>> see the 20ms
>>>>>>> >> diff that you exposed...
>>>>>>> >>
>>>>>>> >> Is there a way to automate the result checking, so it could help
>>>>>>> to monitor
>>>>>>> >> itsp quality
>>>>>>> >
>>>>>>> > Here
>>>>>>> > https://github.com/openpreserve/scape-xcorrsound/issues/17
>>>>>>> > this is a link to a tool which compares two sound files at the wave
>>>>>>> > level. You can also see the sound samples in the ticket.
>>>>>>> > The problem is, that currently this tool does not detect such
>>>>>>> slipping frames.
>>>>>>> >
>>>>>>> > I will also write a detailed article with examples, during the
>>>>>>> weekend.
>>>>>>> >
>>>>>>> >
>>>>>>> >
>>>>>>> >
>>>>>>> >
>>>>>>> >>
>>>>>>> >>
>>>>>>> >> 2015-04-04 12:53 GMT+02:00 Stanislav Sinyagin <
>>>>>>> ssinyagin at gmail.com>:
>>>>>>> >>>
>>>>>>> >>> Important is to send the media through two independent carriers
>>>>>>> - then you
>>>>>>> >>> can be sure that your clock is compared against a high quality
>>>>>>> one.
>>>>>>> >>>
>>>>>>> >>> On Apr 4, 2015 12:46 PM, "Regis M" <
>>>>>>> regis.freeswitch.org at tornad.net>
>>>>>>> >>> wrote:
>>>>>>> >>>>
>>>>>>> >>>> Hi,
>>>>>>> >>>>
>>>>>>> >>>> It's a very interresting test and conclusion.
>>>>>>> >>>>
>>>>>>> >>>> I will try to build a similar environnement on my freeswitch to
>>>>>>> check
>>>>>>> >>>> some system :)
>>>>>>> >>>>
>>>>>>> >>>> thanks
>>>>>>> >>>>
>>>>>>> >>>>
>>>>>>> >>>>
>>>>>>> >>>> 2015-04-04 2:16 GMT+02:00 Stanislav Sinyagin <
>>>>>>> ssinyagin at gmail.com>:
>>>>>>> >>>>>
>>>>>>> >>>>> hi,
>>>>>>> >>>>>
>>>>>>> >>>>> Here are some test results which might be interesting:
>>>>>>> >>>>>
>>>>>>> >>>>> I've set up a test FreeSWITCH server at digitalocean (they use
>>>>>>> KVM),
>>>>>>> >>>>> and originated test calls to some ITSP accounts. Each call
>>>>>>> played a
>>>>>>> >>>>> test sound file (voice samples from ITU:
>>>>>>> >>>>>
>>>>>>> http://www.itu.int/net/itu-t/sigdb/genaudio/AudioForm-g.aspx?val=1000050
>>>>>>> >>>>> ), and on the receiving end another FreeSWITCH server was
>>>>>>> recording
>>>>>>> >>>>> the received session. The receiving server was on a XEN VM.
>>>>>>> Both
>>>>>>> >>>>> servers talked G711 to their corresponding ITSP accounts.
>>>>>>> >>>>>
>>>>>>> >>>>> Then I compared the source audio and received audio in
>>>>>>> Audacity editor.
>>>>>>> >>>>>
>>>>>>> >>>>> The tests showed that the received voice wave shifts 20ms from
>>>>>>> the
>>>>>>> >>>>> original sound every few seconds, randomly back and forward.
>>>>>>> This
>>>>>>> >>>>> skewing was always present, regardless of ITSP and time of day.
>>>>>>> >>>>>
>>>>>>> >>>>> Then I moved the sending server to a XEN VM, and the waves
>>>>>>> appeared
>>>>>>> >>>>> almost perfectly synchronized (there was a 5ms skew in one
>>>>>>> test, and I
>>>>>>> >>>>> guess it's related to transcoding buffer, as the sound was
>>>>>>> obviously
>>>>>>> >>>>> transcoded).
>>>>>>> >>>>>
>>>>>>> >>>>> The effect was always visible with a 30-second sample, and I
>>>>>>> also made
>>>>>>> >>>>> additional tests with 2-minute samples.
>>>>>>> >>>>>
>>>>>>> >>>>> It would also be interesting to test this from a VmWare VM. If
>>>>>>> someone
>>>>>>> >>>>> wants to run such a test, feel free to contact me directly.
>>>>>>> >>>>>
>>>>>>> >>>>>
>>>>>>> >>>>> cheers,
>>>>>>> >>>>> stan
>>>>>>> >>>>>
>>>>>>> >>>>>
>>>>>>> >>>>>
>>>>>>> _________________________________________________________________________
>>>>>>> >>>>> Professional FreeSWITCH Consulting Services:
>>>>>>> >>>>> consulting at freeswitch.org
>>>>>>> >>>>> http://www.freeswitchsolutions.com
>>>>>>> >>>>>
>>>>>>> >>>>> Official FreeSWITCH Sites
>>>>>>> >>>>> http://www.freeswitch.org
>>>>>>> >>>>> http://confluence.freeswitch.org
>>>>>>> >>>>> http://www.cluecon.com
>>>>>>> >>>>>
>>>>>>> >>>>> FreeSWITCH-users mailing list
>>>>>>> >>>>> FreeSWITCH-users at lists.freeswitch.org
>>>>>>> >>>>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>>>>> >>>>> UNSUBSCRIBE:
>>>>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>>>>> >>>>> http://www.freeswitch.org
>>>>>>> >>>>
>>>>>>> >>>>
>>>>>>> >>>>
>>>>>>> >>>>
>>>>>>> _________________________________________________________________________
>>>>>>> >>>> Professional FreeSWITCH Consulting Services:
>>>>>>> >>>> consulting at freeswitch.org
>>>>>>> >>>> http://www.freeswitchsolutions.com
>>>>>>> >>>>
>>>>>>> >>>> Official FreeSWITCH Sites
>>>>>>> >>>> http://www.freeswitch.org
>>>>>>> >>>> http://confluence.freeswitch.org
>>>>>>> >>>> http://www.cluecon.com
>>>>>>> >>>>
>>>>>>> >>>> FreeSWITCH-users mailing list
>>>>>>> >>>> FreeSWITCH-users at lists.freeswitch.org
>>>>>>> >>>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>>>>> >>>> UNSUBSCRIBE:
>>>>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>>>>> >>>> http://www.freeswitch.org
>>>>>>> >>>
>>>>>>> >>>
>>>>>>> >>>
>>>>>>> _________________________________________________________________________
>>>>>>> >>> Professional FreeSWITCH Consulting Services:
>>>>>>> >>> consulting at freeswitch.org
>>>>>>> >>> http://www.freeswitchsolutions.com
>>>>>>> >>>
>>>>>>> >>> Official FreeSWITCH Sites
>>>>>>> >>> http://www.freeswitch.org
>>>>>>> >>> http://confluence.freeswitch.org
>>>>>>> >>> http://www.cluecon.com
>>>>>>> >>>
>>>>>>> >>> FreeSWITCH-users mailing list
>>>>>>> >>> FreeSWITCH-users at lists.freeswitch.org
>>>>>>> >>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>>>>> >>> UNSUBSCRIBE:
>>>>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>>>>> >>> http://www.freeswitch.org
>>>>>>> >>
>>>>>>> >>
>>>>>>> >>
>>>>>>> >>
>>>>>>> _________________________________________________________________________
>>>>>>> >> Professional FreeSWITCH Consulting Services:
>>>>>>> >> consulting at freeswitch.org
>>>>>>> >> http://www.freeswitchsolutions.com
>>>>>>> >>
>>>>>>> >> Official FreeSWITCH Sites
>>>>>>> >> http://www.freeswitch.org
>>>>>>> >> http://confluence.freeswitch.org
>>>>>>> >> http://www.cluecon.com
>>>>>>> >>
>>>>>>> >> FreeSWITCH-users mailing list
>>>>>>> >> FreeSWITCH-users at lists.freeswitch.org
>>>>>>> >> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>>>>> >> UNSUBSCRIBE:
>>>>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>>>>> >> http://www.freeswitch.org
>>>>>>>
>>>>>>>
>>>>>>> _________________________________________________________________________
>>>>>>> Professional FreeSWITCH Consulting Services:
>>>>>>> consulting at freeswitch.org
>>>>>>> http://www.freeswitchsolutions.com
>>>>>>>
>>>>>>> Official FreeSWITCH Sites
>>>>>>> http://www.freeswitch.org
>>>>>>> http://confluence.freeswitch.org
>>>>>>> http://www.cluecon.com
>>>>>>>
>>>>>>> FreeSWITCH-users mailing list
>>>>>>> FreeSWITCH-users at lists.freeswitch.org
>>>>>>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>>>>> UNSUBSCRIBE:
>>>>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>>>>> http://www.freeswitch.org
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> _________________________________________________________________________
>>>>>> Professional FreeSWITCH Consulting Services:
>>>>>> consulting at freeswitch.org
>>>>>> http://www.freeswitchsolutions.com
>>>>>>
>>>>>> Official FreeSWITCH Sites
>>>>>> http://www.freeswitch.org
>>>>>> http://confluence.freeswitch.org
>>>>>> http://www.cluecon.com
>>>>>>
>>>>>> FreeSWITCH-users mailing list
>>>>>> FreeSWITCH-users at lists.freeswitch.org
>>>>>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>>>> UNSUBSCRIBE:
>>>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>>>> http://www.freeswitch.org
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> _________________________________________________________________________
>>>>> Professional FreeSWITCH Consulting Services:
>>>>> consulting at freeswitch.org
>>>>> http://www.freeswitchsolutions.com
>>>>>
>>>>> Official FreeSWITCH Sites
>>>>> http://www.freeswitch.org
>>>>> http://confluence.freeswitch.org
>>>>> http://www.cluecon.com
>>>>>
>>>>> FreeSWITCH-users mailing list
>>>>> FreeSWITCH-users at lists.freeswitch.org
>>>>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>>> UNSUBSCRIBE:
>>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>>> http://www.freeswitch.org
>>>>>
>>>>
>>>>
>>>>
>>>> _________________________________________________________________________
>>>> Professional FreeSWITCH Consulting Services:
>>>> consulting at freeswitch.org
>>>> http://www.freeswitchsolutions.com
>>>>
>>>> Official FreeSWITCH Sites
>>>> http://www.freeswitch.org
>>>> http://confluence.freeswitch.org
>>>> http://www.cluecon.com
>>>>
>>>> FreeSWITCH-users mailing list
>>>> FreeSWITCH-users at lists.freeswitch.org
>>>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>>> UNSUBSCRIBE:
>>>> http://lists.freeswitch.org/mailman/options/freeswitch-users
>>>> http://www.freeswitch.org
>>>>
>>>
>>>
>>> _________________________________________________________________________
>>> Professional FreeSWITCH Consulting Services:
>>> consulting at freeswitch.org
>>> http://www.freeswitchsolutions.com
>>>
>>> Official FreeSWITCH Sites
>>> http://www.freeswitch.org
>>> http://confluence.freeswitch.org
>>> http://www.cluecon.com
>>>
>>> FreeSWITCH-users mailing list
>>> FreeSWITCH-users at lists.freeswitch.org
>>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>>> UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users
>>> http://www.freeswitch.org
>>>
>>
>>
>> _________________________________________________________________________
>> Professional FreeSWITCH Consulting Services:
>> consulting at freeswitch.org
>> http://www.freeswitchsolutions.com
>>
>> Official FreeSWITCH Sites
>> http://www.freeswitch.org
>> http://confluence.freeswitch.org
>> http://www.cluecon.com
>>
>> FreeSWITCH-users mailing list
>> FreeSWITCH-users at lists.freeswitch.org
>> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
>> UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users
>> http://www.freeswitch.org
>>
>
>
> _________________________________________________________________________
> Professional FreeSWITCH Consulting Services:
> consulting at freeswitch.org
> http://www.freeswitchsolutions.com
>
> Official FreeSWITCH Sites
> http://www.freeswitch.org
> http://confluence.freeswitch.org
> http://www.cluecon.com
>
> FreeSWITCH-users mailing list
> FreeSWITCH-users at lists.freeswitch.org
> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
> UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users
> http://www.freeswitch.org
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20150409/a25b92b0/attachment-0001.html

Join us at ClueCon 2016 Aug 8-12, 2016
More information about the FreeSWITCH-users mailing list