[SR-Users] RTPEngine QoS statistics and leg labelling

[Richard Fuchs]

On 2018-02-14 12:21 PM, Alex Balashov wrote:
> Hi Richard,
>
> On Wed, Feb 14, 2018 at 08:18:11AM -0500, Richard Fuchs wrote:
>
>> On 2018-02-14 02:44 AM, Alex Balashov wrote:
>>> Okay, so after having a spelunk through the source code, I have the
>>> impression that this MOS / quality analysis functionality is limited
>>> solely to feedback off RTCP supports.
>>>
>>> I was under the impression from the documentation that RTPEngine does
>>> some stream analysis of its own, although in fairness, the documentation
>>> does not explicitly make this claim or the contrary one way or another.
>>> :-)
>> It does gather some stats about the media streams on its own (even more so
>> with the extended transcoding capabilities now), but at least one component
>> of the MOS calculation (latency) depends on RTCP being provided by the RTP
>> clients, so without that, no MOS score can be produced.
> Which statistics can RTPEngine compute over the stream on its own? Since
> I saw no metrics whatsoever for the B leg, I assumed none. I was running
> mr5.5.3 however. Does this differ in 6+?

It keeps track of packet and octet counts. With transcoding engaged for 
a call, it also keeps track of packet loss and duplications. It could 
calculate jitter, but doesn't currently do so.

>
> Also — I'm sure the question has been asked before — can RTPEngine send
> its own RTCP sender/receiver reports?
>
> My testing setup was:
>
>     Phone --> Kamailio+RTPEngine --> Kamailio+RTPEngine --> Carrier
>                 (where tinkering        (ITSP
> 	        was happening)
>
> Is there a way to turn on backward RTCP feedback from #3 back to #2?
>

Not currently, but we may offer that capability in the future, 
especially to support video in an RTP/AVP to (S)AVFP bridging scenario.

Cheers