From: Robert Elz <kre(_at_)munnari(_dot_)OZ(_dot_)AU>
| Both IPv6 and ATM-to-the-desktop-and-replacing-IP
| - require major changes to applications, hosts TCP code, and the boxes
| that connect hosts.
The IPv6 changes range between fairly minor and non-existant, depending upon
the application (and when done, result in code that still works over IPv4).
To the extent that is true, it could also have been true of ATM. You
could have had yet another address family, gethostbyname2(), and
N-to-A/A-to-N for ATM that would have been just as hard or difficult for
applications as as the IPv6 solution. I suspect whether the application
changes for IPv6 are major or minor depends mostly on whether you're making
them, and whether you have source so that you can make them. If your
application never directly handles addresses, and you have inet_pton()
that works (the widely distributed code is, well, imperfect even for IPv4),
then the changes are not too painful. If your application or worse your
protocol need to deal directly with addresses, then things get sticky.
128 bits pegs don't fit in 32 bit holes, at least not on the systems and
networks I fight.
| - claim to be desert toppings AND floor waxes
No, IPv6 claims only to be a (really fairly minor) enhancement of IPv4
Have you seen the SNL skit?
Whether IPv6 is a really fairly minor enhancement of IPv4 is as true of
ATM to the desktop. The ATM mavens claimed that jacking up desktops and
replacing IP with ATM would have been trivial. On the other hand, some
IPv6 advocates never stop beating the drums for IPv6 QoS, multi-homing,
and so forth and how the Millennium will arrive with IPv6. I think
the truth in both cases is not so simple as either minor enhancment
or dawning of a new age.
ATM requires new hardware (everywhere), IPv6 runs just fine on ancient
hardware (my main stable IPv6 node is a 486 from about a decade ago...
Yes, that's a very good point. Dealing with cells at reasonable speed
couldn't be done with software, although modern routers aren't doing
badly with 48 byte TCP-LW/IP datagrams.
The two aren't comparable at all in terms of what they're attempting to do,
or what is involved in upgrading from one to another.
You have a point in the hardware hassles, but not about the other half.
ATM was driven mostly by telephant hopes and fears, but it also claimed
some of the IPv6 wonders. IPv6 is only rationally justified as a modest
but necessary enhancement to IPv4, but never say that or worse, mention
obvious implications of that fact in the hearing of an IPv6 adovocate or
you'll feel as if it were 1985 and you'd mentioned TCP/IP within earshot
of real (i.e. ISO OSI) network expert. Sean has been provactive, but some
of the reactions have been comically similar to what I heard from
participants in the then dominant standards committee not that many years
ago. In other words, the IETF has come of age.
Vernon Schryver vjs(_at_)rhyolite(_dot_)com