My suggestion is that you schedule a call or meeting with your MPLS provider, your firewall provider, your phone system vendor, and anyone other than you who is responsible for network configuration.
Broad overview:
VoIP has two distinctly different flows that need to be treated differently.
- SIP (or MGCP, etc.) signaling. These are call setup and teardown messages, hold/resume, busy lamps, and other call control messages. This traffic wants an "assured forwarding" class of service. If the packet needs to be delayed for half a second it isn't a big deal but make sure it is delivered because it is used to set up a call. Don't worry about speed of delivery but if you have to make a choice delay rather than drop.
- RTP voice traffic. This is the actual speech. These packets are small but there are a lot of them and it is critical that they be delivered in order and at a steady pace. If there is unavoidable congestion it is better to drop a packet than delay it. A syllable of speech delivered half a second late is worse than not being delivered at all. It wants to be in a class called "expedited forwarding". Put these packets first but if you have to make a choice, drop rather than delay.
QoS mechanisms can be based on a number of things and this is why the MPLS provider, phone system vendor, firewall vendor and you (or if not you the person responsible for router and switch configuration) need to be on the same page. Some things used for decision making:
- Layer 2 COS bits. This is part of the Ethernet header, not concerned with IP addresses or ports. Eight possibilities (3 bits) 0 through 7. Switches typically use this for QoS decisions. Limited granularity with only 8 possibilities.
- Layer 3 DSCP. This is part of the IP header, 64 possibilities (six bits). This is much more granular and here you can specify the difference between expedited and assured forwarding, etc.
- Layer 3 manually specified address/port/protocol. Here you identify traffic based on the specifics of the packet such as source/destination IP and port and protocol. Typically done at the entrance edge if the connected device isn't capable of encoding DSCP.
Once you gain knowledge of the mechanism used by the phone system, the MPLS network, and the firewall(s), then it's a matter of ensuring that the QoS is classified correctly, marked appropriately for each device and/or network, and that the forwarding decisions honor the marking. Also necessary is reclassifying and/or stripping QoS tags applied to traffic that your management policy decides are not to be given priority (such as online video games).
Then it's a matter of configuration and testing. You shouldn't need to replace your Adtran gear, it's a matter of setting things up correctly and keeping them that way in the event of moves/adds/changes. I'm assuming that the corporate office has higher bandwidth than a T1 which is only good for about 20 simultaneous G.711 calls assuming no other traffic.
