Re: IPv6 and the "average user"

Janos Farkas (chexum@bankinf.banki.hu)
Thu, 21 Nov 1996 13:08:39 +0100 (MET)


On 20 Nov 1996, H. Peter Anvin wrote:
> Quite true, but the IPv6 address space is expected to be exhausted in
> 30 years. The switch from IPv4 to IPv6 is going to be painful, but
> that is nothing compared to what switching standard protocol stacks in
> the late 2020's would be like...!

Which brings up a question; could we make a framework, which could allow
changing protocols under applications, without recompiling?
I.e. currently to change an application to support IPv6 we need to hack
around, and change socket(AF_INET,..) into a conditional
socket(AF_INET6,..) and changing the resolver calls to something more
modern, (and possibly this may care about getting the address family for
us too in the socket() call). And to care about the "port" abstraction
being the same, or being accessible by getservby..(). And to hope all the
stream/dgram semantics are sufficiently close. Or maybe the NRL
getaddrinfo() is about to solve this? :)

I imagine that someone has a single ftp program, which is compiled once
with our c library, and it is able to do connections to tcp/ipv4 and
tcp/ipv6 (that's already almost normal.. :), AND tcp/ipx, spx/ipx, maybe
TP/OSI, and anything one can think of. It looks to me a goal which is
worth to think about.

Janos