We run a Call Manager cluster with centralized call processing. Branch offices (connected via T1) are currently in the same city and have the phones at the branch offices register to the centralized CCM cluster over the T1. This works great.
We are beginning to add branch offices across the nation. We added one office with a site-to-site VPN over the Internet and we are having issues with dropped calls (not sure offhand if its signalling or voice packet related, but I think it's signalling related). 2 different Internet providers are currently used.
The nation-wide telecom vendor that we are consolidating to has an all fiber MPLS network with very low latency, jitter and virtually no packet loss. The metrics seem ideal for voice. They claim that if we peer directly to them for Internet access on both sides that we should able to use an Internet VPN (with QoS implemented on our borders) for voice until we get to the point where we need to utilize VPLS for multipoint between our geographic sites. We will only be traversing their all fiber MPLS network.
What is the best way to configure the branch offices? Should we do a site-to-site VPN and let the phones register to the centralized CUCM cluster and use SRST for fallback? Or, should we use CME with a SIP trunk so the phones can register locally? We'd prefer the first method so everything can be centrally managed.
I've found docs online regarding centralized processing with branch office configurations that show the phones register over the "WAN", but can't find any documentation that outlines the latency and jitter metrics for voice signalling and traffic over the "WAN". We don't want to commit to a 12, 24 or 36 month T1 and continue to have issues.