Ah, well... there will be an extra service charge to have Stuxnet installed. Check with the sales office for more details. I think they are running a "2-for-1" promotion -- THIS MONTH ONLY! :>
Remember, UDP's "eficiency" (if you want to call it that) comes at a reasonably high cost!
There are no guarantees that a given datagram will be delivered (or received). The protocol that you develop *atop* UDP has to notice when stuff "goes missing" (e.g., require an explicit acknowledgement, sequence numbers, etc.)
There are no guarantees that datagram 1 will be received *before* datagram 2. "Turn off plasma cutter" "Move left" can be received as "Move left" "Turn off plasma cutter" (which might be "A Bad Thing" if there is something located off to the left that doesn't like being exposed to plasma! :> )
There is no sense of a "connection" between the source and destination beyond that of each *individual* datagram. Neither party is ever aware if the other party is "still there". (add keepalives if this is necessary)
There is no mechanism to moderate traffic as it increases (and, those increases can lead to more dropped/lost datagrams which leads to more retransmission *requests*, which leads to more traffic which leads to... keep in mind any other traffic on your network that could worsen this -- or, be endangered by it!)
Appliances other than switches can effectively block UDP connections. If you ever intend to support a physically distributed domain that exceeds what you can achieve using "maximum cable lengths" (one of the drawbacks about moving away from "orange hose" and its ilk was the drop in maximum cable length), you have to be careful in considering what any *other* "interconnect appliances" do to the traffic you intend to pass (and, if your protocol will be routed!)[It's surprising how *short* "100m" is when it comes to cable lengths! Esp in a manufacturing setting where you might have to go *up* a considerable way -- and, leave a suitable service loop -- before you can even begin to go "over"! And, line-of-sight cable routing may be impractical. For example, here (residential), the *average* length of a network cable is close to 70 feet -- despite the fact that the (2D) diagonal of the house is just about that same length *and* all the drops are centrally terminated!]
When someone later decides it should be a piece of cake for your "engineering office" to directly converse with your controllers located in the "manufacturing facility", you'll find yourself explaining why that's not easily accomplished. "Why not? I can talk to Google's servers in another *state*/country... (damn consultants always trying to charge extra for stuff they should have done in the first place!)"[N.B. Raw ethernet frames don't even give you the above (lack of) assurances :> ]