Danny had a very big influence on the split - he had a great
analogy between milk and wine - it was ok for wine packets
to be delayed (they got better with age) but milk was something
else. That led to UDP for real-time stuff, especially voice
packets - yeah, we were doing packet voice as far back as 1975.
vint
At 11:21 PM 7/17/2001 -0700, Stephen Casner wrote:
On Tue, 17 Jul 2001 mstjohns(_at_)mindspring(_dot_)com wrote:
Vint spoke on this very topic as the dinner speaker for the TCP/IP
Interoperability Forum in Monterey (the pre-Interop interop
conference). As I recall from his comments, the decision to split
the two came about as a mutual epiphany between him and Dave Clark
as they were trying to figure out some way around some of the
stupidity in NCP.
I believe Danny Cohen was one of the primary instigators for the
split, motivated by the desire to support packet transmission of
voice. I recall a presentation Danny made in which he talked about
TCP's reliability function R which was not appropriate for
delay-sensitive voice communication. He suggested that we could apply
another layer to implement an R^-1 function, but that it might be more
efficient to split TCP so that a different (transport) protocol could
be implemented without the reliability function.
-- Steve