ietf
[Top] [All Lists]

Re: An alternative to TCP (part 1)

2001-02-12 08:20:03

As long as it is safe (i.e., follows good congestion control),
why should we care how many of these protocols are defined?  After
we ensure the protocols are safe we can just let Darwinism take its
course.

Because a customer with sufficient $ would inevitably request the
Frobnitz Transport, since it garnered the IETF imprimateur and the
code base would eventually collapse under the weight of all this
"innovation".

Maybe I spoke too quickly above.  The points I was trying to get at:

  * TCP isn't necessarily the answer for every application.
    
  * We don't need to try to make TCP (or TCPng) the answer for every
    application.
    
  * Maybe ALF is the way to go...  But, I am not convinced that we
    cannot provide transports that go a long way towards solving a
    good chunk of the problems, such that most developers don't need
    to develop their own protocols on top of UDP due to their
    application's needs.

  * So, I don't see a need for a TCPng, per se.  But, if folks want
    to write new transports that include different mechanisms (e.g.,
    no strict ordering requirement) I think that is wonderful. 

  * I like the SCTP model (build something new) better than the
    model where we try to develop TCPng (rework something old, make
    sure it still works with the old stuff, etc.).  This just seems
    like The Way To Go at the transport layer to me.

allman


---
Mark Allman -- BBN/NASA GRC -- http://roland.grc.nasa.gov/~mallman/