[ih] The origin of variable length packets

[dave.walden.family at gmail.com]

On Mar 1, 2011, at 4:25 AM, John Day <jeanjour at comcast.net> wrote
> But it also occurred to me that this was not such a big deal and was probably discovered by everyone who went to build one.
> 
It certainly seemed obvious given we were implementing it primarily in software.
I can imagine that when Baran wrote his paper he was thinking about an all hardware implementation.  I can also imagine that in those pre ARPANET days they were thinking about some much less flexible multiplexing approaches.

My impression (maybe stated in Larry  Roberts' paper where the Heart et al.paper  was presented) was that the ARPANET could happen when it did because the economics and technology of general purpose computers and communications circuits had reached appropriate crossover points.  That let packet switching  (e.g., by having variable size packets, leased circuits, dynamic routing, store-and-forward by IMPs, etc.) provide an approximation of good performance for bulk transfers (which previously had been the specialty of message switching -- open the circuit when you have enough traffic to keep it busy for a while and then close the circuit) and an approximation of highly interactive communication (which had previously used circuit switching -- open the circuit and keep it open until you are done talking back and forth).

The early voice experiments over the ARPANET required the "raw packets" capability of the IMPs to be enabled which forgot about reliable (more or less), ordered transmission of messages of packets across the net ( which slowed up speech too much) and delivered single packets in whatever order they got to the destination IMP.  At least this is how I remember it. John Makhoul would remember more. (Unfortunately Jim Forgie died recently.)

>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>> 
>>