mark@cbosgd.UUCP (Mark Horton) (01/31/84)
peter honeyman: With this view, domain-ing *is* just another way of routing (a view i took every opportunity to express in dc). There are indeed many similarities. The key difference is that routes are relative to the sender, domains are relative to the root of the tree, so your domain based address is always the same no matter where the sender is. And, of course, the delivery software need not traverse the exact path down the domain tree to get the message to you, there are often better ways to get it there. However, I'm unclear how this fact fits into what you're saying, I think it's similar. we now get into the whole domain naming farce (now playing in your nearest newsgroup). the geographical boundaries of domains are a red herring; the real issue is the quality of a registry's routing tables. so here's my proposal (in domainist notation): let's have a UUCP domain (what the heck), and invent a few other domains called HARPO, ULYSSES, LINUS, and the rest of the backbone sites, sites where mailaholics keep the quality of local tables high. (local means whatever you like -- we're talking operational view here -- a site is local if the path to it is known.) backbones then route mail through the various domains. This sounds amazingly similar to what we have in mind. The key difference is that, while the subdomains probably would be built around backbone sites, they would be given names like NJ (New Jersey) and NE (New England), rather than HARPO and LINUS, so that people would not have to change their published mailing addresses when a backbone site decided it was somebody elses turn. (The exact name is up to the subdomain to decide, of course, they would not have to be geographic like these examples. I tend to think in geographic terms, but there is no reason to restrict domains to a geographic organization.) (this should be recognizable as standard uucp routing; you'll forgive me if i say harpo!cbosgd!mark instead of mark@d.OSG.CB.ATT.HARPO.UUCP.) Where you've left off the uucp! from the front, much as people might like to leave off the .UUCP from the end of a domain address? The address uucp!harpo!cbosgd!mark looks a lot like mark@cbosgd.harpo.uucp, doesn't it? I hope that if you always type harpo!cbosgd!mark, you won't insist that your mailer route it through harpo even though there's a better link through allegra or directly (assuming it can be deterimined that the cbosgd is the same machine). i kidded a domainist in dc that the only way domains would take hold in the usenet world would be to integrate them into netnews. what i'm now proposing is that we take the best of usenet -- it's connectivity -- and build on that. Funny, there has been support in netnews for domains since last July. Of course, netnews is not a mail transport mechanism, so all it has to do is generate, preserve, and present domain addresses to the mail program, it isn't up to netnews to figure out how to get it there. Mark Horton
honey@down.UUCP (02/07/84)
i was hacking pathalias tables the other day, messing around with the
wjh12!bitnethost%user vs. psuvax!user@bitnethost.BITNET controversy.
it wasn't that hard to solve -- i think it was something like
BITNETDOMAIN = @{...}(DIRECT)
BITNETSITES = {...}%(DIRECT)
BITNET BITNETDOMAIN(0)
psuvax .BITNET
wjh12 BITNETSITES
i treat BITNET exactly like a site name, which is consistent with
my suspicion about domains: they are site names in disguise. for
example, domainists talk about a domain registry, doing name serving or
some other jargony thing. i don't know how a domainist views the
concept of "BITNET-ness", but for me it suffices to conceive of whatever
site does the trick for bitnet. (according to my tables, that's
psuvax. and i don't lose sleep worrying how psuvax will handle mail to
psuvax!BITNET!root.) with this view, domain-ing *is* just another way
of routing (a view i took every opportunity to express in dc). BITNET
is psuvax (or any site with a bitnet table), ARPA is any site that will
gateway to arpaland, MILNET is ... etc. the only hassle is UUCP, where
route tables are ad hoc.
we now get into the whole domain naming farce (now playing in your
nearest newsgroup). the geographical boundaries of domains are a red
herring; the real issue is the quality of a registry's routing
tables. so here's my proposal (in domainist notation): let's have a
UUCP domain (what the heck), and invent a few other domains called
HARPO, ULYSSES, LINUS, and the rest of the backbone sites, sites where
mailaholics keep the quality of local tables high. (local means
whatever you like -- we're talking operational view here -- a site is
local if the path to it is known.) backbones then route mail through
the various domains. (this should be recognizable as standard uucp
routing; you'll forgive me if i say harpo!cbosgd!mark instead of
mark@d.OSG.CB.ATT.HARPO.UUCP.)
i kidded a domainist in dc that the only way domains would take hold in
the usenet world would be to integrate them into netnews. what i'm now
proposing is that we take the best of usenet -- it's connectivity --
and build on that.
peter honeymanhenry@utzoo.UUCP (Henry Spencer) (02/08/84)
I agree with pretty much everything Mark said, except for one thing:
.............I hope that if you always type harpo!cbosgd!mark, you won't
insist that your mailer route it through harpo even though there's a
better link through allegra or directly (assuming it can be deterimined
that the cbosgd is the same machine)....
Damn right I insist on it. If I wanted the best route, I'd type something
like "mark@cbosgd". It is absolutely essential to have some way to
bypass brain-damaged idiotic "user-friendly" optimization when necessary,
and people who make their mailers optimize exclamation points out deny
this vital facility to anyone who routes mail through their site. I am
starting to feel sufficiently strongly about this that I'd be tempted to
suggest that any site that insists on optimizing whether it's asked to
or not should be disqualified from being a backbone site, on the grounds
that its software is too "user-friendly" to be usable.
Pursuing this point further, to avoid having this message be just a
repetition (albeit with different emphasis) of stuff I've said before:
if we are going to implement domain-based addressing, we *MUST* *MUST*
*MUST* implement the extra bit of syntax that provides for explicit
routing. With any luck, wide availability of domain-based software will
help avoid further brain-damaged "cleverness" on mailers that should be
following orders and not trying to outsmart their customers.
"I don't want a user-friendly system: I'm not a friendly user."
--
Henry Spencer @ U of Toronto Zoology
{allegra,ihnp4,linus,decvax}!utzoo!henry