In a message of  Rick Van Ruth (3:640/937.6) wrote:

Hi Rick,

 DF>> One thing that I couldn't clearly see though, you need an IP address
 DF>> for each _interface_ rather than an IP address for each machine.  So,
 DF>> if your Win95 machine uses an ethernet you have an IP address for
 DF>> that ethernet.  Then your Amiga should have an IP address for it's
 DF>> ethernet and _another_ IP for it's PPP dialup.  Any routing
 DF>> statements and the like must take this into account.

 RVR> Ahh! Here's this point again. There seems to be 2 groups of thinking
 RVR> when it comes to TCP. One group, to which you seem to belong,
 RVR> believes that your interface needs a separate IP, the other group do
 RVR> not support this theory.

 RVR> I belong to the "other" group :)

Then you're in the wrong group.  

 RVR> Logically speaking, your interface is not an address, so why does it
 RVR> need an address allocated? You are basically saying I need 4 IP's for
 RVR> my 2 machines. True this will work but it is not really needed. The
 RVR> computer is the address, it is what processes the TCP packets and
 RVR> protocols.

Logically speaking, each it's the interface that has a network connection
so it's the interface that _should_ have an address.  Ummm, how to simplify
this so that I don't end up typing an essay on the subject...  ummm... oh
well let's try it and see where we end up...

You have a local area network using ethernet interfaces and using TCP/IP
networking.  Each machine on that network must have it's own unique IP
address in order for TCP to work properly.  TCP uses those IP addresses to
route data around and see to it that it ends up in the right place.

You have a wide area network (connection to the internet) using a PPP
interface and using TCP/IP networking.  Once again, each machine on that
wide area network must have it's own unique IP address so that data ends up
where it's supposed to end up.

Now the key point, each of these networks is separate.  If a single machine
is going to exist in both of these networks then it is going to have to
have relevant IP address information for each network.

If it only has an IP address for the WAN, how is the LAN supposed to know
how to send it traffic?  If it only has an IP address for the LAN, how is
the WAN supposed to route traffic to it?

This is why an IP address really refers to the INTERFACE rather than the
computer.  In the above case, one machine is going to have to have two IP
addresses, one that relates to it's existance in the LAN (an IP address for
it's ethernet interface) and one that relates to it's existance in the WAN
(an IP address for it's PPP interface).

So, the final part is that it's the routing statements on the machine that
has both interfaces that is important in getting all the machines on your
LAN able to use the WAN connection.  This single machine has to be able to
understand how to take traffic moving over your LAN that requires data from
a machine not on your LAN and move it over to the WAN so that it gets where
it has to.  It then has to be able to do the reverse, take traffic from the
WAN and route it back to the right machine on the LAN.

 RVR> I look at it as if I only had a ppp link to my ISP then I would need
 RVR> 2 addresses, one for the Amiga and one for the PPP. But this is not
 RVR> the case, the ppp is the same IP as my computer, it is given (in the
 RVR> state of dynamic connections) by the ISP and is the address used to
 RVR> contact any TCP related programs on my machine.

Wrong.  If you only had a PPP link to your ISP and nothing else then you'd
only need one IP address, the one for your PPP connection.  If you add in
an ethernet connection on that same computer then you'll also need a second
IP address that refers to that other network that also happens to be
running on that one machine.

The address that is allocated by your IP will certainly work for any
protocols that are running over the connection between you and your ISP but
that same IP address will not be suitable for using on your local area
network as, basically, your LAN won't know how to deal with it.

 RVR> Also, my ISP (and a few others I have seen) run extra machines for
 RVR> their own purposes off their server, this is usually done with
 RVR> ethernet cards. They only give the machine connected an IP, they
 RVR> don't allocate addresses to the actual ethernet connection other than
 RVR> to give it the machines IP.

I've yet to see an ISP that doesn't run multiple machines on a LAN of some
sort.  When these machines are setup they only have one IP address because
the only network connection is the ethernet interface. If, on the other
hand, you look at the terminal server (be it a computer with multi-serial
cards or a dedicated box like a Cisco or some such) you'll find that it has
quite a few IP addresses associated with it.  In effect, it will have an IP
address associated with each serial line it's dealing with (which is the IP
address that you will be assigned when you connect to it), plus an IP
address associated with it's existance in relation to the serial ports plus
another one associated with it's existance on a LAN (ethernet) and perhaps
even another IP address depending on how that box connects to the internet
itself (if it has built-in ISDN support for example).

I can't think of an easy way for you to test this as it requires a fair
degree of knowlege of the IPS's layout to start delving into his IP
addresses but if you talk to your local friendly ISP about it - and if they
really are that friendly about it - they might be willing to show you over
their setup and you'll see this sort of thing in action.

  //  CYA,
\X/ Dave ;-)

... Good intentions randomize behavior.         

---