In <[log in to unmask]> [log in to unmask] writes:

<warning: long winded explanation follows... ;-) >

> > We are running a Novell 4 Network of PCs, with our HP3000 Connected. We a> r
> > trialing a
> > POP3/SMTP Package on the HP3000 which works great with our W95 PCs using
> > Outlook.
> >
> > We have a parallel email system using Lotus cc:Mail and I'm pre-empting o> u
> > requirements
> > and saying we may need to bridge the two. Since we have no DNS on the IP
> > Network, how will
> > the mail routing be performed. I presume you need DNS to get the recipien> t
> > machine's IP address
> > so you know which node to forward undelivered mail onto - or am
> > misunderstanding the way
> > Internet email works here...I'm relatively new to all this Networking
> > stuff...<g>

Mark,

  Mailers need to resolve the hostname portion of an e-mail address to a
machine (or list of machines) somewhere. The typical method of doing this
is by calling upon a DNS (domain name server), passing it the hostname in
question, and getting back a list of systems to try in order to get mail
through to the intended recipient. The particular "DNS" records for this
purpose are "MX" or mailer-redirect records. Failing access to a name server,
a host may also just simply "look up" the host itself via a local network
configuration file (like /etc/hosts on Unix systems, HOSTS.NET.SYS on a
3000, the NMMGR directory on a 3000, or in our case -in addition to the
others- you can define local hostname->IP address mappings in our mail
system). For "real" e-mail service on the Internet, you have to have at
least one machine with access to Internet-aware DNS (as opposed to the
inside-firewall/internal DNS some organizations run). For e-mail around
inside an organization though, any of the above resolution methods will
work just fine. [3000 to 3000 email using our package will even take
advantage of HP3000s ability to "probe" the local network to dynamically
to find another HP3000 (or HP9000) on the wire without any need for NMMGR
or other directory entries at all.]

  For instance, say you want to send mail to [log in to unmask] The
mailer will call upon a DNS server to get a list of mail servers that will
"accept" mail for compuserve. In some cases it may be only one machine on
the list; in compuserve's case, you get a list of about 20 machines.
  These machines are "sorted" by a "preference" number; the lower numbered
entries being the ones a mailer is supposed to try first. From there, the
mailer goes "down the list" trying to talk to the mail servers on those
machines til it (1) gets the message through, (2) gets a definitive error
(like "no such person here"), or (3) runs out of hosts to try.
  In the event of (3), most mailers will queue the message and attempt again
later (due to the nature of the Internet, (3) is a common occurrence and the
SMTP mail transfer protocol was designed with this in mind).

> > DNS has always puzzled me I'm afraid. AFAIK, it's just a way to translate>
> > name to an IP
> > address, but how does the machine requesting the conversion know where th> e
> > machine
> > with the "database" is? How does this work on the Internet? Is there a
> > giant server some-
> > where with all the domain names and IP addresses on it? <g>

Mark Bixby explained the global DNS service in his earlier message quite
well, so I'll leave that as is. I will enthusiastically add that since he did
recently port *the* major name server package (BIND) to the HP3000, that you
can now run your *own* DNS server right on your 3000. (BIND isn't a trivial
package to use -and especially to debug if you have a complex network- but
setting it up for a simple network is pretty easy.)

      Hope this helps, (and I didn't put TOO many people to sleep)
           Chris Bartram