[evla-sw-discuss] Delay models to station boards

[Barry Clark]

Kevin has a valid point that we don't know whether datagrams will work
adequately without a test.  However, I don't see how to arrange a 
realistic test without having the whole correlator room networking in
place.  (Although actually, I am more concerned that datagrams might 
get lost within the mchost stack than I am about the network.)  What 
should we do to test deliveries?   (At a minimum, we should put a sequence
number in the document.)

It's not totally clear that connected sockets help - having a document
delivered late is in some ways worse than not at all.