i agree that this is a nervy thing, and i don't like the way it's currently
done, but unless i am missing something, i don't see how adding ANY information
to a documentt will fix this.
> > to do that without such a hack. Multiple IPs per host _is_ a hack. It can
> > only be done on a small number of OSes, and it is a real waste of IP
yes, it's a hack, a a benefit of running a real os (where this ability is there
exactly for this reason - hopefully a temporary solution until...)
> The internet draft draft-gulbrandsen-dns-rr-srvcs-00.txt may, with one
> change (an additional "actual port" data field in the RR), be used to
> solve this "problem". But it would require all the clients to change,
> every single one.
>
.. the above can be implemented. doing this by dns is the correct way (analogous
to the correct use of MX records to do similar functional vis-a-vis email).
> I'd appreciate comments on the draft; should I put back in "actual port"?
> It was there in the very first version, but went away for lack of
> perceived use.
>
if this change to dns is being considered, that's good enought for me - year
ago you couldn't get any to even listen to such an idea.
as for the current hack being clever - it ain't; ip alias games has been a useful
unix admin tool for years - saying it's clever is like saying "ain't it clever
to put a bandage on that cut" - it's obvious.
as for the apache hint - unless something has changed, how to do this is not
documented, and the actual routing manipulations will probably be somewhat
unix vendor/setup specific; there's at least two ways this can be done.
rich