In message
<200004261332(_dot_)JAA01189(_at_)hygro(_dot_)raleigh(_dot_)ibm(_dot_)com>,
Thomas Narten typed:
IPv6's claimed big advantage - a bigger address space - turns out
not to be an advantage at all - at least in any stage much short of
completely deployment.
Not surprisingly, I disagree.
right, noels wrong.
the amount of address translation
state you have to keep (and syncronise between failover NATs etc)
per active session decreases as
the percentage of hosts that are native IPv6 increases,
(and obviously also
decreases as the absolute number of hosts increases but new hosts
are all v6) - in your scenario (a likely wireless 3G deployment one,
this could happen pretty fast
the amount of disconnect the v4 legacy machines will see because the
state maintenance will fail (as any large system does partially, ALL THE TIME),
will increase, and possibly quite fast...routing state is already in a
bad enough state....without adding address translation state to it...
NATs are not only bad e2e karma, they are bad tech, just like x.25 and atm.
Here's why:
>> if you have a site which has more hosts than it can get external
IPv4
>> addresses for, then as long as there are considerable numbers of
IPv4
>> hosts a site needs to interoperate with, *deploying IPv6 internally
to
>> the site does the site basically no good at all*.
Actually, in the above scenario, NAT is already a requirement for IPv4
communication with the outside world. So, if you switch to IPv6
internally, use IPv6-IPv4 NATPT (i.e., combination of NAT and IPv6 to
IPv4 translation) you have pretty much the same
functionality/limitation as with IPv4 NAT.
Now, consider someone in the process of deploying massive numbers of
devices (100's of millions) together with the infrastructure to
support them (e.g., wireless). With IPv4, they face not only the
necessity of using NAT to get to outside destinations, but also the
use of NAT _internally_ because there isn't enough private address
space to properly number the internal part of the infrastructure.
In this scenario, IPv6 internally at least gives them end-to-end ness
internally (plus scalability, more robustness, etc., etc.), something
the can't get with IPv4. And it gives them the same set of
issues/headaches when talking to the outside world that they would
have if just using IPv4.
I don't know about you, but it scares me to read the various forecasts
about how wireless will transform the landscape over the next few
years. E.g., more wireless phones with internet connectivity than
PCs. The numbers are just staggering and the associated demand for
addresses will be astonishing. We ain't seen nothing yet.
Thomas
cheers
jon