Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!usc!ucsd!hub.ucsb.edu!spectrum.CMC.COM!lars
From: lars@spectrum.CMC.COM (Lars Poulsen)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: Internet Access Costs
Message-ID: <1990Nov29.212559.7284@spectrum.CMC.COM>
Date: 29 Nov 90 21:25:59 GMT
References: <315@srchtec.UUCP> <1990Nov26.233848.20828@nmt.edu>
Organization: Rockwell CMC
Lines: 55

In article <315@srchtec.UUCP> mra@srchtec.uucp (Michael Almond) writes:
>>One other item: PSI says we will not have a constant IP #.  It supposedly
>>changes each time we establish a connection with them.  However, our address
>>will remain constant (probably searchtech.com).

In article <1990Nov26.233848.20828@nmt.edu> rmilner@zia.aoc.nrao.edu
   (Ruth Milner) writes:
>I fail to see how this can be the case. Granted they can change their tables
>any time they want, *but* they have the following restrictions:
>
>     a) they have to use an Internet number assigned to them/you by the NIC
>     b) the rest of the Internet has to understand the routing (if it's
>        advertised, and if it isn't no-one can reach you)
>     c) your system has to know *before establishing the connection* what
>        its own Internet address is
>    ...
>Even if it *could* be done, it would be such an administrative nightmare
>trying to figure out "who's 123.456.789.10 this morning?" that I can't think
>why they *would* do it.
>
>Ruth Milner
>Systems Manager                     NRAO/VLA                    Socorro NM

Ruth,
	The network number stays constant, the IP address changes.
(The confusion seems to have been caused by a bad choice of words:
The posting says "internet number" for "IP address" and "address" for
"fully qualified domain name".)

	Crazy as it seems at first glance, dynamic IP addresses are
not actually all that unusual. Client-only PC's often do this. And with
domain name service, it is doable even to handle incoming connections.

	Yes, it does create a few management hassles. You need to have a
name server with a transaction interface to change the name/address
mappings on the fly (where the run-of-the-mill BIND just reads a table
when it is reloaded) and you cannot have management utilities that
assume that name/address mappings are constant.

	But technically, at the low level, it is no harder than ARP.
You need an ARP-like handshake before you switch to IP mode. Most SLIP
interfaces already have a handshake (unix login !) before turning into
IP interfaces, so there definitely is a mechanism to do that.

	As to WHY one wants to do this ? When you have a couple hundred
PC's around, it is easier to just declare them to be generic than to try
to keep track of locations and owners for all of them. Likewise, network
providers may have only allocated class C network numbers for a routing
cluster, and may prefer the dynamic addresses over changing backbone
routing implementations with hardcoded netmasks.

/ Lars
-- 
/ Lars Poulsen, SMTS Software Engineer
  CMC Rockwell  lars@CMC.COM