VoIP packet mis-insertion between ATA and E&M card

[nathan.dixon]

Hi,

I have run some subjective audio tests between an ATA and and E&M card in a 3745 router and found problems end to end. ATA to ATA and E&M to E&M perform perfectly. We have narrowed the problem down in a lab by having a single ethernet crossover between the ATA and 3745. Still we see problems. When we look at call parameters we see late and early packets plus a lot of Predicted seconds. We expect to see some but there are a lot on the ATA to E&M setup which we think are causing the subjective tests to fail. This is not unit related as we have tried many units and upgraded to latest H323 ATA firmware and latest IP plus on the 3745. I am sure QOS can't come into this as this is an isolated setup with 100M/full duplex etherenet between the two devices. The predicted seconds parameter starts to run up after about 60-90 seconds of call time and after a few minutes maybe about 500mS in total. This does not happen to the same extent with the ATA to ATA or E&M to E&M, maybe 50mS after 30minutes! Is this buffer related or a Cisco 'feature'!? Any help appreciated.

Regards,

[Chester Rieman]

Hi Can you please define where you are seeing the 'predicted seconds' parameter increasing...what command was used to see this?

Can you put a sniffer in the middle here and look at the packets. Ethereal provides RTP analysis to help define what is happening. Do you see anything unusual on the ethernet interfaces when you do a show int?

Try this to see direct DSP counters.

Weh the call is up, do a 'sho voice call summ'

you should see the active call up as S_CONNECT.

Note the port number for this call in the first column and now enter 'sh voice call x/x.x.x' where the x'x are the port number from the previous command, you should get dsp statistics output that shows the stats straight from the DSP (must have term mon enabled if telnetted to the gateway)

[nathan.dixon]

Hi,

We are seeing the predicted seconds increase on the E&M card using the "sh call activ voice" command. All usul Fastethernet interface stats look ok.Will have to try the sho voice call xxx later this week as I'm not near the lab today

See attached "sh call activ voice"

Thanks for the response.

[Chester Rieman]

I don't see 'predictive seconds' in the output but here is where I do see an issue:

RxOutOfSeq=9<--main issue,packets out of sequence

RxLatePak=0

RxEarlyPak=1

PlayDelayCurrent=55

PlayDelayMin=55

PlayDelayMax=65

PlayDelayClockOffset=-293476173

PlayDelayJitter=0

PlayErrPredictive=180<--causes this

PlayErrInterpolative=0

PlayErrSilence=0

PlayErrBufferOverFlow=20

PlayErrRetroactive=0

PlayErrTalkspurt=0

So do a sniff at the ATA and a sniff at the router, are the RTP packets in sequence when they leave the ATA and out of sequence when they arrive at the router? If so, network issue. If they are out of sequence leaving the ATA, please save the capture and open a case with TAC.

[nathan.dixon]

Hi,

Looks like I phrased it wrong "GapFillWithPrediction=180 ms" is the actual term. But i guess this is only the result of the out of sequence or lost packets. The network cannot be at fault as there is a crossover cable directly between the two devices.

Will get my packet sniffer setup!

Thanks for the response.