"Erik E. Fair" writes:
> Anyway, this is not going away, and given that, I still think that a
> hostmask for the NetBSD network code is a good idea if we want to
> support a lot of IP addresses on the same interface efficiently.

I agree with you that if the problem was simply one of handling the
craze for having one machine have 30 ip addresses each with a
different "www.foobar.com" name, your solution would be
ideal. Unfortunately, we have other applications.

For example, take the IPv6 world. In IPv6 land, a host typically has
many addresses (I say "has" rather than "will have" because
implementations are now interoperating). An IPv6 host has a local link
address, several mandatory multicast addresses, global addresses for
every prefix associated with the network (a network will have multiple
prefixed during renumberings), etc. Admittedly the v6 network stacks
are different code, but its nice to be able to share as much as
possible. There is also the problem of the use of virtual hosts to
emulate hosts that have "gone" -- having a new machine take over some
critical service like DNS for a local network after the original
machine has gone but before all the resolvers and domain servers and
the like can be updated. I've done this a bunch over the years in
emergencies.

Were the problem just to have a host emulate 32 "virtual" hosts I
agree that your proposal would be the right thing, but given this
other long term requirement I suggest that we find a good solution for
all applications, like hash tables.

Perry