reid@glacier.ARPA (Brian Reid) (11/23/85)
Various combinations of events have forced me to spend the week or two of hacking that was needed to get our machine to call itself "glacier" in all uucp contexts. Our world is quite case-insensitive, and both "glacier" and "Glacier" have been used in the past. I'm pretty sure I now have all of the case-folding code installed in all of the relevant places (you wouldn't BELIEVE how many places it matters) so that we can continue to be Glacier on our local network, GLACIER.STANFORD.EDU on the Arpanet, and glacier on UUCP. Down with the 4.2BSD "hostname" mechanism! Machines can have different host names on different networks! -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
honey@down.FUN (Peter Honeyman) (11/24/85)
what's the advantage of having different host names on different networks?
zben@umd5.UUCP (11/25/85)
In article <629@down.FUN> honey@down.FUN (Peter Honeyman) writes: >what's the advantage of having different host names on different networks? 1. Politics. Our IBM/BitNet machines have UMD<letter> names: UMDA thru UMDD. Our Internet machines have UMD<digit> names: UMD2, UMD7, etc. So, what should the name of our Sperry (which is on both nets) be? We respond to the names: UMDU.BITNET and UMD2.UMD.EDU both.* 2. Routing. If your machine is called AAA.BITNET and BBB.XXX.EDU then (assuming the remote side has enough smarts to handle two name spaces without .<netname> suffixes - UMD2 does this) you can control which path mail from a particular person uses by giving him one name or the other. This is really off-the-top-of-the-head stuff, and I would welcome any other reasonable surmises why one would want to do this. It would help to have a mantra of them to chant when I have to look at the kluge mail code... Ben Cranston POSTMASTER@UMD2 * Although (again due to politics) we also respond to UMD2.BITNET. Don't ask. It's a REAL kluge since BitNet doesn't have aliases. We sort of lie to UMDA and tell it that there are *two* machines down that wire. There are also masses of kluges at UMD2 to handle the equivalence of the names (AND the fact that other BitNet sites don't know this equivalence). It also breaks a lot - i.e. my kluges are not yet quite robust enough. Also note there used to be a UMDU.ARPA alias back in the days of .ARPA names but it seems to have gone away at some point... -- Ben Cranston ...{seismo!umcp-cs,ihnp4!rlgvax}!cvl!umd5!zben zben@umd2.ARPA
jsdy@hadron.UUCP (Joseph S. D. Yao) (12/01/85)
The only reasons to have multiple names for a machine are:
(1) politics, (2) a bad technical implementation of name
handling (that can't handle the "true" name of the machine),
(3) politics, and (4) politics.
Machines started calling themselves foo.ARPA and fum.BITNET
as part of the domain concept. The domain concept, however,
has nothing whatsoever to do with routing. The idea behind
the domain concept (which you should read in RFC 92?, rather
than taking my word for it) is that each machine should have
a unique name within its subdomain, which ... which should
have a unique name within its own domain. Any machine may
have a routing table for any other machines, despite number
or lack of common subdomains. If there is no routing for
the machine, then the sending machine should have a routing
for a routing machine in one of the domains or subdomains,
which latter machine should then be able to route messages
down to the recipient machine.
In other words, UMDU.UMD.EDU is a reasonable name whether
you're routing a machine over BITNET or ARPANET or whatnot,
according to the domain concept. Any name conversion needed
by the hardware should be done by the software.
Realities include the fact that the domain concept is not
fully implemented, and many sites still do not conform to it
for technical or political reasons or just from lack of
understanding (or caring) about the domain standard. I
should also mention that not everybody agrees what it means
or that it is good, although most people who have a grasp of
the naming problem agree that It Is a Problem, and Something
Should Be Done About It.
--
Joe Yao hadron!jsdy@seismo.{CSS.GOV,ARPA,UUCP}mark@cbosgd.UUCP (Mark Horton) (12/02/85)
In article <113@hadron.UUCP> jsdy@hadron.UUCP (Joseph S. D. Yao) writes: >The only reasons to have multiple names for a machine are: >(1) politics, (2) a bad technical implementation of name >handling (that can't handle the "true" name of the machine), >(3) politics, and (4) politics. There is more to it than just politics. One obvious problem is that the "true" name of a host has to be unique, and this uniqueness implies a fairly long name, like Glacier.Stanford.EDU. To the locals, the machine is no doubt just called "Glacier", and they view that as the "true name". So partially qualified names are one good reason for a difference. Another is nicknames. The machine houxp.UUCP is locally called the "purple machine", and the p in the name stands for purple. Lots of places have nicknames or even more local abbreviations like "A". A third is that UNIX has yet to deal well with the notion of a machine name. System V has the notion of a "nodename", which is limited to 8 characters (6 or 7 in some contexts) and 4.2BSD is oriented toward a single name like "glacier", both with the assumption that the higher levels in the domain aren't included. Heirarchical name spaces are still a fairly new thing and the tools are a bit green. The standards for getting at such local names (which are critical) don't exist at all. And then there is history. At one time, the same machine at Berkeley was known as "ucbvax" (on UUCP), "CSVAX" (on the Berknet), and "Ernie" (spoken.) It's only a feat of luck that it wasn't also the machine known as "Berkeley" (on the ARPANET.) For reasons of upward compatibility, such names stick around for quite a while. Which brings up another reasons: different gateways. At one organization, the gateway onto ARPANET may be a different machine from the gateway onto UUCP which is different from the gateway onto BITNET. While in the ideal world the name of the gateway would have nothing to do with the name of the domain, in practice domain and gateway names are often chosen so the same name can apply to both the domain and the gateway, witness "Berkeley" as a machine on the ARPANET and "titan.JUNET" as the domain name for the Tokyo Institute of Technology. Finally, there is the question of whether "Glacier", "glacier", and "GLACIER" are the same name. In the ARPA world they are. In the UUCP world there is considerable pressure to use only lower case, due to implementations that treat cases differently. In the human world at Stanford, the nice upper case first letter is important for display purposes. It is unfortunate that this issue even came up, but it did. Mark