[VoiceOps] network jitter tools

[Mark R Lindsey]

On Aug 25, 2009, at 9:32 PM, Peter Childs wrote:

>
> Passive taps will only give you 'partial' view of performance,  
> unless both ends support RTCP in which case life is much better.


RTCP is very useful for some things, but not for jitter. Estimating  
jitter certainly APPEARS easier with RTCP -- but the facts disappoint.

You can have a RTCP-reported jitter, but still have a terrible phone  
call.

The RTCP-reported jitter average masks periods where the jitter buffer  
overflows (due to router queue compression) or underflows (due to long  
delays). The jitter calculation in RTCP gives, at best, a vague long- 
term statistical view.

For example: using the RTCP jitter statistic defined in RFC 3550,  
suppose you have 20 ms ptime RTP, but network congestion causes each  
packet to arrive 25 ms apart. Suppose also you have a 6*ptime (120 ms)  
jitter buffer that begins playback when it's filled to 3*ptime (60  
ms). RTCP will dutifully report 5 ms of jitter, but in reality the  
user will experience alternating periods of audio and silence (The  
pattern will be something like: 75 ms silence followed by 600 ms  
audio, then 75 ms silence, then 600 ms audio, etc.)

RTCP XR (RFC 3611) does provide some additional useful data on this in  
the discard-rate field. This is a direct measurement of jitter buffer  
behavior. However, I believe you can still have silent periods due to  
network congestion, but report no discards and trivial jitter averages.





mark r lindsey at e-c-group.com http://e-c-group.com/~lindsey +12293160013