What is this guy smoking? Send a UDP packet from one node and receive it on another. Exactly-once delivery. If what he really means is "guaranteed to be delive

What is this guy smoking? Send a UDP packet from one node and receive it on another. Exactly-once delivery. If what he really means is "guaranteed to be delivered exactly once", then, yeah, no, of course not, because if your network goes down, you can't send anything. Over a network. You can, of course, just have one big fat machine and have two nodes in the machine and send the packet from the one node to another in the one big giant machine. There's no "network" to go down, unless somebody slices into the machine with a red-hot katana. You can also have a fully-connected topology where every machine is hard-wired into every other machine, and every machine will route for every other machine. There's no real "network" to go down; as long as there are 2 hosts that aren't dead, a message from one will get to another. What I find amusing is that well before you get into truly hard distributed system problems, for a sufficiently complex implementation/application, your apps will be riddled with so many bugs and so many operational problems that the thing is going to fall over way before a well-built network goes down.