| given TCP's inability to do congestion-sensitive backoff during
| connection establishment, I'm not so sure that increasing the number
| of concurrent connection attempts is a good idea.
I am pretty sure it is NOT a good idea.
| the selection of the server with the ... lowest [RTT]
| was often ... good but rarely ... optimal.
One of the things which is becoming apparent to a number
of service providers is that that a low RTT is necessary
but insufficient for "user happiness" - RTT _stability_ is
often more important.
Regrettably, there is little to no way for an end system
to understand the short term stability of a path other than
by trying to move traffic across it.
| no, I can't see how a routing-system-based multihoming
| will scale down to SOHO networks either.
Why not? It should be an explicit goal of the next cut
at an IPng, and shouldn't just stop at SOHO -- it should
go right to individual homes, _at least_. I have multiple sets of
wires coming into my flat, and live in a regulatory environment
where each of those wires (and some new ones) is allowed to carry
data/POTS/ISDN/cabletv/you-name-it.
And then there's Sweden, which is several steps ahead.
Check http://www.bredbandsbolaget.se/eng/node103.asp?ID=16
Yes, they really are serious about more than a quarter of a million
homes which will have the moment-to-moment chance to use
any or all of: FTTC/FTTB/FTTH from these folks, cable modems,
or the various offerings from the various telcos (xDSL, dialup)
for Internet connectivity.
The days when only a handful of middle-aged Swedish women have broadband
connections like this: http://www.stupi.se/Bilder/7296-3251-0759/g.html
into their apartments are numbered. (BTW, if you look closely, you'll
see her apartment's network was multihomed already.)
The assumption that only larger organizations will want multihoming
and at-will-provider-changing is a bad one.
Sean.
