[cisco-voip] Call quality stats on SIP trunk?

[Robert Kulagowski]

On Thu, Apr 25, 2013 at 1:04 PM, Wes Sisk <wsisk at cisco.com> wrote:
> Robert,
>
> My CUBE contact says:
>
> "sh call act voice is where is look first. Looking at the output below I do
> not
> see much diff. in Tx/Rx packet counts for the two leg. So that looks good.
> DSP stats are not relevant as for this call CUBE isn’t even going to touch
> the
> rtp payload (no xcoding/MTP in use.)
> A sniffer at the egress leg will give us lot in terms of evidence of what we
> are receiving and putting out on the wire.
>
> What exactly is the voice quality issue here? CPU checks out ok ?"

So now I've had my first "call sounded bad" issue with the SIP trunk,
but I want to make sure I understand what I'm looking at.

The leg that points at CM had this when the call ended:
P-RTP-Stat: PS=41299,OS=6607840,PR=33182,OR=5309120,PL=0,JI=0,LA=0,DU=826

Packet loss, jitter and latency are all 0. Great! But not.

The problem is that from the perspective of the phone / user, it
sounded bad. Pulling the stats from CM shows that the call had a MOS
of 2.6, and the user complained of it sound like it was "underwater".
PS (CUBE) is 41299, PR (Phone) 31139, loss of 10160 packets from
CUBE->Phone
PR (CUBE) is 33182, PS (Phone) 41315, loss of 8133 packets Phone->CUBE

So, plenty of packet loss. But how come the PL stat is "0"? Why isn't
specify something like PL-S and PL-R? Because the phone certainly
thinks there was packet loss, but that's not reflected in the
end-of-call stats on the CUBE.

Until I know why the discrepancy exists, I don't know that I can trust
the stats on the outbound call leg showing me "perfect" calls.

(But I also need to determine why I'm losing so many packets on the
inside network.)