[ILUG] Asterisk / Trixbox Weirdness

[Brendan Minish]

Hello 

I have spend the last two days getting nowhere with the following
issue. 

I have been helping out with trying to resolve an issue with a call
centre that is having issues with dropped calls 

they are using a fractional E1 PRI provided by BT on Eircom's copper 

the issue is as follows. When a call comes in it is answered and the
caller is played (Music on hold) whilst asterisk sets up a SIP call with
one of the call agents that are in the pool for the Queue. 

Normally this all works fine and the caller is transferred from
Listening to the Music on hold (actually a ringtone ) to the agent that
has just picked up the call.

However sometimes (about 1 call in 10) at the point where the call gets
transferred to the agent the line goes dead. the call stays up on the
PRI until the caller hangs up(*) but audio does not pass in either
direction between the agent and the caller. This seems to happen a bit
more often when several calls are in progress 

(*) at which time the agent sometimes hears a fraction of a second long
burst of audio from the caller 

Asterisk is version 1.4.21-1 as distributed with trixbox CE 

the PRI cards we have tested are a sangoma (with hardware echo
cancelling) and a Zaptel 
the phones are predominantly snom 360's with the latest firmware 

We do not see this issue at all on internal sip calls and it does not
appear that calls that are taken from the Queue successfully and then
transferred to another agent get affected 

We can't by any means rule out a Provisioning error by BT or line
degradation as the Zaptel (but not the sangoma) seems to drop the entire
PRI on rare occasions 

The issue first arose at around the time that another 100 numbers were
added to the DDI pool

things we can rule out 

1/ it's not an interrupts going missing issue, not one single missed
interrupt is happening, ever 
2/ Hardware has not changed and we are not running into resource limits
(Dual core Xeon with 2Gb ram ) The server handles no other services. we
have tried apic=off and also normal apci. 
3/ IP network, phones are on a separate switch fabric with ample
capacity for VOIP
4/ it's very unlikely to be an issue with the phones.
5/ it's not heat related 
6/ Prior to my involvement a complete bare metal reinstall of trixbox
was attempted as a resolution for the issue. 

It does occur to me is that it may perhaps be related to the PRI since
it always occurs at the point where the call is switched from 'Music on
Hold' a wav file playback to an agent. This must require timing to be
'bang on' or the frame is lost 
The Sangoma and the Zaptel derive their timing from the PRI line and we
are possibly seeing line related issues, of course BT tell us the line
is perfect and Eircom say 'it's nothing to do with us'  
the very short burst of caller audio that sometimes happens as the
caller hangs up seems to indicate that the SIP connection from the
agent's phone to the call was made correctly.

On the other hand it looks like a software issue since surely for Music
on hold to work the framing has to be ok and the D channel signalling
has set up the call correctly.. 

Has anyone ever met anything like this before?
 Any debug suggestions? 
Anyone got details on how to configure sangoma or zaptel cards for
Eircom/BT PRI's in Ireland (for comparison purposes) 

We are under pressure to get this resolved by Tuesday AM and it's
beginning to really spoil the bank holiday weekend. 

regards
.brendan