Introduction
I have limited SIP/Mitel experience, but my networking is pretty strong. The PBX was already configured before I joined so have limited knowledge of how Mitel works, but I have read the deployment guide for networking requirements.
The Problem
We have recently opened a branch office and have deployed some Mitel & Polycom handsets (2 of each, 4 in total). The audio drops for around 15 seconds, usually both ways after 4-6 minutes into a call, internal or external, teleworker mode or native mode.
Teleworker mode works fine from other sites, and the handsets at the primary office do not suffer this issue.
Configuration
LDN3300 hosted at primary data centre as virtual appliance
Primary office:
Has a L2 link (400Mbps) to the data centre where the LDN resides (separate voice VLAN stretched across sites)
We use Juniper SRX switches & SSG firewalls (though voice does not touch this firewall)
We have around 300 handsets working fine here, no issues and around 50 teleworkers - not sure on concurrent calls, but no issues present
New site (Branch Office):
L3 link to ISP, 50Mbps
IPSEC VPN to datacentre where the PBX is - all VLANs allowed, minimal firewalling (voice VLAN at branch allowed to full voice VLAN at DC)
~50 computers
4 handsets
Juniper SRX switches (x2, virtual chassis) , SSG firewalls (x2, active/passive)
Network configuration - separate voice VLAN, DHCP issued by server with 125 option, LLDP to auto place phones into voice VLAN - this works as expected. Phones connect through VPN.
Voice VLAN has priority set and correct frame forwarding in line with Juniper's recommendation (I think it is assured forwarding).
Troubleshooting Performed
Teleworker mode - same symptoms as native
Disabled ICMP redirect on switches & firewall (though none were captured during a port mirroring session)
Bandwidth monitoring - not in contention during audio cut out
Latency - very steady 3-5ms ping from computer to the PBX. Occasional spike up to 20ms, but often during audio cut out it can maintain ~5ms and not fluctuate
Switch monitoring - low utlisation, low CPU/mem, port buffers not exceeded
Firewall - disabled ALG (Application Level Gateway for SIP - in my experience this never plays nicely with anything)
QOS - cannot be set end to end to due L3 link and VPN
iPerf - UDP test consuming 5Mbps (much more than should be consumed for 4 simultaneous calls) from computer in voice VLAN at branch office to a server in the voice VLAN at DC - no packets were lost, jitter at 0.7ms,
Port mirroring of the phone to capture in wireshark shows UDP packets sent back and forth during the good part of the call, but during the audio drop it only shows outbound from the phone - nothing coming in, which explains the silence, and presumably these aren't being received at the other end (not confirmed).
The LDN Call Statistics during a bad call shows high packet loss of ~25% at max burst and total, jitter buffer underflow is 10000-15000
For the good part of the call, avg jitter is 2ms, 0 packet loss, jitter buffer underflow is 10
Does anybody any further ideas to investigate?
