With VoIP, packet loss and delay eventually make the service useless. So VoIP fails harder than either traditional IP apps and PSTN.
That is only in theory. In practice, during times of impending congestion collapse, IP network operators reconfigure the network to cope. For instance when DDoS is detected, people set up ACLs and trigger black hole routes. I think that it is possible for network operators to define an analogous action plan to stave off congestion collapse in an emergency situation. I'm not sure exactly what that action plan would look like, but I'm sure other list members will have plenty of good ideas. If you'll recall, just a few days ago people were talking about how they informally identified IP connectivity to emergency response sites so that those sites could be given priority in restoring service. We just need to sit down and talk these things over with our local emergency response organizations and learn where network operators can become part of the solution.
On the other hand, in a circuit switched network you can do all kinds of interesting stuff (such as restarting all your control software) without breaking your sessions. We're only now seeing this in IP, and I think it's not really possible to reach the same levels with IP routing even in the long run.
MPLS may have the edge here because you can have backup paths and fast reroute to keep traffic flowing if you have an orderly plan for rebooting routers.
And voice over any kind of packet infrastructure introduces significant additional delays.
Experience with the Inter-NOC phone system http://www.pch.net/inoc-dba/ seems to suggest otherwise. Some kinds of packet infrastructure only introduce insignificant delays. It would be interesting to know if any of the academics among us have studied the behavior of a SIP-based VoIP network during various types of failure and congestion scenarios. I suspect that problems will be mostly found under certain specific sets of conditions and if we know what those conditions are and how they impact voice services, then we can plan actions to mitigate the problems. One thing that IP network operators can do is throw bandwidth at a problem by "shedding load", i.e. killing traffic that is deemed non-essential. This would free bandwidth for traffic that is deemed "important". This has nothing to do with QoS per se becaus it can be implemented in many ways up to and including unplugging sites that generate non-essential traffic. All indications are that the next few decades will see an increased number of emergency situation like the tsunami, terror attacks in major cities, hurricanes, earthquakes. We have gotten very good at running the network through normal times, maybe we should now focus on how to keep it running through times of extreme stress. --Michael Dillon