[BXXPwg] Draft Minutes of last week's meeting

Marshall Rose mrose+mtr.netnews@dbc.mtview.ca.us
Wed, 9 Aug 2000 21:12:10 -0700


> I did wonder if this might be specified on a per-profile basis, so that
not
> every application would have to code for the more complex case if they
> don't (directly) use profiles with piggyback data.

right now, you get to specify with each profile what "initialization"
elements may be present.

you can use these either as "constructors" or as a way to reduce network
latency.

i think there seems to be general concensus that beep is never going to be
the most efficient protocol on the wire. of the various things to optimize a
protocol for (scalability, efficiency, simplicity, robustness,
extensibility), you need to make trade-offs. in the beep case, these each
get a nod, although some nods are much more vigorous than others.

however, there are different kinds of efficiency. the one where you usually
get the biggest payoff is in reducing network latency (because that's
independent of how fast your hosts are). if you look at the way the SASL and
TLS profiles get used (once, at the start of the session), it's pretty clear
that if you can cut in half the number of round-trips, you can dramatically
reduce the time it takes before you get to do useful work, regardless of
what the hosts are. and that's why i like piggybacks.

/mtr