Right. This sort of thing is one of the recurring themes on the
end2end-interest list; every now and then some bright spark announces
a shiny new protocol that outperforms TCP, and is gently reminded that
if everyone used it the results would be much worse than with modern
TCP implementations.
We _really_ don't want a repeat of the meltdown that originally prompted
Van Jacobson's work on TCP congestion avoidance. FHTTP, and also the HTTP
modifications proposed and analyzed in
http://www.ncsa.uiuc.edu/SDG/IT94/Proceedings/DDay/mogul/HTTPLatency.html
seem to be headed in the right direction; multiple parallel TCP connections,
on the other hand, are pretty disturbing.
Thomas Maslen My opinions, not Verity's
[email protected]