> Right, but if you have a CIDR network then the broadcast address is an
> unknown... You could just send the ICMP to the gateway and hope it's
> implemented (is it required by any standards?), but I don't think
> relying on a broadcast is a good idea.

Not sure what everyone else was thinking, but I consider these broadcast
techniques to be proposals for an additional option "probe network to set
defaults". In any other form, I would have serious reservations about them.

As a very general rule, algorithms that blatantly trust outside state SHOULD
be strictly optional, but if that is not possible then they MUST be documented
thoroughly, and should produce well designed error messages.

One of the most ridiculous things about Microsoft O/S's is how they have
incredibly smart default settings yet there's almost no way to find out what
decisions they actually made. So it's almost impossible for _anyone_, let
alone Microsoft customer support, to diagnose problems quickly and accurately.

I am all in favor of discussing algorithms for divining some of the network
parameters from the surrounding network. But these MUST NEVER be a substitute
for auditable and editable manual settings. If anything, the network prober
should be a separate program that people can run as a cheap diagnostic tool.

Todd Whitesel
toddpw @ best.com