Often times it's mission critical systems where latency is key, for many reasons: - you have a human in the loop to take a split second decision, every millise

Often times it's mission critical systems where latency is key, for many reasons: - you have a human in the loop to take a split second decision, every millisecond counts - you have a very short time to perform 'looped' operations - where the result of one measurement must be taken into account to effect the next measurement (adaptive optics, some radar systems, some mechanical control loops) and you can't wait. You'd think 'oh but you got more than 1 ms for that' Well not always since one must take into account the time to detect the failure, and the time to switch other parts of the system (which sometimes must be done in sequence with the connection takeover). I'd say 'forget tcp' there but we don't always get to decide the comm layer...