[Freeswitch-users] Choppy audio when conferencing 4+ participants

Michael Jerris mike at jerris.com
Tue Apr 18 23:59:39 MSD 2017


It is certainly possible to run in a vm if done correctly, its just done incorrectly more than correctly.

> On Apr 18, 2017, at 3:50 PM, Bipin Patel <bipin at xbipin.com> wrote:
> 
> Just to add my experience, I have a windows server running as VM on a dedicated server hosted with iweb, they call it smart server and I have been using freeswitch on it from a long time now and everything works perfect audio wise so I guess this has something to do with Linux and the kernel run by Amazon on their VM.
> 
> Correct me if I'm wrong
> 
> On April 18, 2017 8:58:00 PM Anthony Minessale <anthony.minessale at gmail.com> wrote:
> 
>> In general when a thread about performance or vm turns to lots of theories.  This is a science so we need facts to diagnose and its often not possible once we get this far down the rabbit hole.  To the contrary, asking about timer test was a good start!
>> 
>> 
>> 
>> 
>> On Tue, Apr 18, 2017 at 11:41 AM Michael Jerris <mike at jerris.com <mailto:mike at jerris.com>> wrote:
>> Superstition comment was in regards to tc malloc, not you.
>> 
>> 
>>> On Apr 18, 2017, at 7:35 AM, David Ponzone <david.ponzone at gmail.com <mailto:david.ponzone at gmail.com>> wrote:
>>> 
>> 
>>> Anthony,
>>> 
>>> Could you elaborate a little bit more on why I am being superstitious ?
>>> Is timer_test command obsolete ?
>>> Or is it a way to emphase the fact that you won’t support FS on VM, anyway ?
>>> 
>> 
>>>> Le 17 avr. 2017 à 23:27, Anthony Minessale <anthony.minessale at gmail.com <mailto:anthony.minessale at gmail.com>> a écrit :
>>>> 
>> 
>>>> I believe we may be stumbling into superstition at this point.
>>>> 
>> 
>>>> On Fri, Apr 14, 2017 at 6:03 PM, David Ponzone <david.ponzone at gmail.com <mailto:david.ponzone at gmail.com>> wrote:
>> 
>>>> Did you run the timer_test command ?
>> 
>>>>> Le 14 avr. 2017 à 23:04, Bilal Dar <bilal at rgate-systems.com <mailto:bilal at rgate-systems.com>> a écrit :
>>>>> 
>>>>> The issue triggered when I ran out of inodes on the server, even after freeing inodes things never went back to normal. I was running earlier  m3.large and now moved to m3.2xlarge servers, CPU/memory utilization is negligible. 
>>>>> 
>>>>> Model   vCPU    Mem (GiB)       SSD Storage (GB)        
>>>>> m3.large        2       7.5     1 x 32  
>>>>> m3.2xlarge      8       30      2 x 80  
>>>>> 
>>>>> 
>>>>> 
>>>>> On Fri, Apr 14, 2017 at 1:53 PM, Colin Morelli <colin.morelli at gmail.com <mailto:colin.morelli at gmail.com>> wrote:
>>>>> Robert,
>>>>> 
>>>>> While I'd love to see VMs provide more stable ground for FS, it's simply not the best task for a VM. Virtual machines scale well specifically because you can overprovision them. It would not be nearly as cost-effective to run VMs if each instance had a guaranteed dedicated slice of hardware to operate on. While hypervisors are very good at task prioritization, they're not perfect. If the hypervisor can't schedule processor time when FS needs it because the CPU cores are momentarily taken on other tasks, there's not a whole lot FS can do. This is not an issue with just FS, but with all real-time applications. In most apps, even large clock skews and bad hypervisors schedulers can go completely unnoticed. If there's consistent 5-10ms every time you click to load a web page, you'd probably have no idea. If there's 5-10ms jitter every time you try to read 20ms of audio, you have really bad audio. Granted most skews are not that bad, but the effects are pronounced when you're dealing with data that's real-time in nature.
>>>>> 
>>>>> Bilal,
>>>>> 
>>>>> I have no idea what AMI you're running, but a very rough "ear test" has made me fairly confident that I can get better performance running AmazonLinux AMIs over Ubuntu (and probably many others). It wouldn't surprise me if AmazonLinux builds a custom kernel that has been tuned to run better on AWS hardware. I'd say it's at least worth a quick experiment.
>>>>> 
>>>>> Best,
>>>>> Colin
>>>>> 
>>>>> On Fri, Apr 14, 2017 at 4:34 PM, Mundkowsky, Robert <rmundkowsky at ets.org <mailto:rmundkowsky at ets.org>> wrote:
>>>>> Michael <>
>>>>>  
>>>>> 
>>>>> Just curious, why so many problems with VMs?
>>>>> 
>>>>>  
>>>>> 
>>>>> I would think most applications need real time clocks that provide consistent valid data?
>>>>> 
>>>>>  
>>>>> 
>>>>> Bilal,
>>>>> 
>>>>>  
>>>>> 
>>>>> you might try a larger AWS instance to make sure your are getting 100% of the box; might help some.
>>>>> 
>>>>>  
>>>>> 
>>>>> Robert
>>>>> 
>>>>>  
>>>>> 
>>>>> From: freeswitch-users-bounces at lists.freeswitch.org <mailto:freeswitch-users-bounces at lists.freeswitch.org> [mailto:freeswitch-users-bounces at lists.freeswitch.org <mailto:freeswitch-users-bounces at lists.freeswitch.org>] On Behalf Of Michael Jerris
>>>>> Sent: Friday, April 14, 2017 4:25 PM
>>>>> To: FreeSWITCH Users Help <freeswitch-users at lists.freeswitch.org <mailto:freeswitch-users at lists.freeswitch.org>>
>>>>> Subject: Re: [Freeswitch-users] Choppy audio when conferencing 4+ participants
>>>>> 
>>>>>  
>>>>> 
>>>>> can you reproduce the same issue on real hardware?  We’ve seen all kinds of weird timing issues that could account for this running on aws.
>>>>> 
>>>>>  
>>>>> 
>>>>> On Apr 14, 2017, at 4:16 PM, Bilal Dar <bilal at rgate-systems.com <mailto:bilal at rgate-systems.com>> wrote:
>>>>> 
>>>>>  
>>>>> 
>>>>> Its an AWS m3.2xlarge instance.
>>>>> 
>>>>>  
>>>>> 
>>>>> On Fri, Apr 14, 2017 at 9:57 AM, Michael Jerris <mike at jerris.com <mailto:mike at jerris.com>> wrote:
>>>>> 
>>>>> Real hardware or VM?
>>>>> 
>>>>> > On Apr 14, 2017, at 12:48 PM, Bilal Dar <bilal at rgate-systems.com <mailto:bilal at rgate-systems.com>> wrote:
>>>>> >
>>>>> > I have been struggling with an issue for almost 2 weeks.
>>>>> >
>>>>> > Our regular calls have no quality issue and looking RTCP statistics network conditions are perfect. We have normally on peak hr 60 calls and around 10 conferences.
>>>>> >
>>>>> > We have noticed that when we have 2 conferences of 4 or 5 participants, audio starts breaking for the users who are on conference. Regular calls do not experience any quality degradation.
>>>>> >
>>>>> > I upgraded the server to specs of 30Gig memory and 8 vCPU but still the issue exists. Common thing I have noticed even during off-peak hrs is that two 4+ participant call can cause the issue.
>>>>> >
>>>>> > I have ruled out network & hardware. Last change I made was moved all users to G.711 from G.722. Now I am not sure what other steps I can take. Appreciate any suggestions.
>>>>> 
>>>>>  
>>>>> 
>> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20170418/e74b8d16/attachment.html 


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