smart@ditmela.oz (Robert Smart) (10/25/89)
And of course it works the other way. MX-registration allows a host with a proper domain name to be reached from the Internet. If the Internet wants to be reachable from other networks they have to go and ask those other networks to put in a gateway. For example the Internet could ask ACSnet's minder to put in a gateway from ACSnet to the Internet so that (for example) an ACSnet user can send mail by sendfile -amailer user@host.university.edu <message and it would magically work just as if host.university.edu was a real part of ACSnet. This could be made to work. Perhaps the Internet is too BIG and IMPORTANT to ask other networks to put in gateways. In which case there seems to be a certain asymmetry in arrangements. Actually I don't see the problem. "user%node" is a perfectly legitimate "local part" of an RFC-822 address, and I don't see why I shouldn't set up my RFC-822 compliant domain to have that local part mean that the mail should be forwarded to some local host that is too insignificant to register. Certainly we shouldn't mandate that meaning of "%". If other RFC-822 compliant domains want to have actual users with %s in their names then that is also ok. Bob Smart <munnari!ditmela.oz.au!smart> or <smart%ditmela.oz.au@uunet.uu.net> P.S. I think I'll stick to %-hacked address in my signature until my faith in the Domain System goes up a lot.
barmar@kulla (Barry Margolin) (10/25/89)
In article <7696@ditmela.oz> smart@ditmela.oz.au (Robert Smart) writes: >Perhaps the >Internet is too BIG and IMPORTANT to ask other networks to put in >gateways. In which case there seems to be a certain asymmetry in >arrangements. Are these other networks using standardized protocols? Yes, there's an asymmetry. The Internet is an order of magnitude larger than most other networks. >Actually I don't see the problem. "user%node" is a perfectly legitimate >"local part" of an RFC-822 address, and I don't see why I shouldn't >set up my RFC-822 compliant domain to have that local part mean that >the mail should be forwarded to some local host that is too insignificant >to register. Certainly we shouldn't mandate that meaning of "%". If >other RFC-822 compliant domains want to have actual users with %s in >their names then that is also ok. No one has proposed legislating away the % hack. It can't be done, for the reason you state: every host is free to interpret the local part as it sees fit. They're just trying to encourage people to switch to the better solutions that have been devised since % was invented. The % hack has several problems: 1) Users have to remember gateways. smart@ditmela.oz.au is easier to remember than smart%ditmela.oz.au@uunet.uu.net. 2) % isn't the only such hack, and it's not always clear what happens when multiple hacks are used together. What is the meaning of foo!bar%baz@quux? Is it quux -> foo -> baz -> user bar or quux -> baz -> foo -> user bar? 3) Problems occur when network topology changes. If uunet.uu.net is replaced as the Internet->UUCP gateway, all the %host@uunet.uu.net addresses are invalidated. This means that users must learn a new gateway, and mailing lists all over the place have to be updated. The domain system simply requires updating the database, and users never notice. This can be very important if the topology change is temporary (for instance, uunet goes down for a week because the company is moving to a new location, so someone else takes over gatewaying in the interim). 4) The domain facility has additional features. For instance, there can be multiple MX records for a host, so that mail can still go through when the default forwarder is down. With %, the decision of which forwarder is used is made statically by the sender. 5) You're dependent on an arcane, nonstandard feature being implemented on a host over which you have no control. Uunet may decide at any time to stop supporting %, and you'll run into problems. Besides encouraging hosts to switch away from %, I think they are also encouraging new systems NOT to implement it in the first place. More importantly, it sends a signal to implementors not to try related kludges. For instance, I've heard of systems that try to parse non-local local parts, in an attempt to optimize routes. They see foo%bar@baz, realize that they know a host named "bar", so send the message directly to bar rather than routing through baz; this loses if baz doesn't interpret % as they think it does, or if baz has a different notion of the host bar. Barry Margolin, Thinking Machines Corp. barmar@think.com {uunet,harvard}!think!barmar
smart@ditmela.oz (Robert Smart) (10/25/89)
In article <31038@news.Think.COM> barmar@kulla (Barry Margolin) writes: > > Are these other networks using standardized protocols? Who's to say that protocols that have been dragged through a long tedious standardization process are better than those created by a few hackers in a back room. And which category does TCP/IP fit into. Certainly TCP/IP standards are not registered with the Standards Association of Australia, unlike some others I could name. > > Yes, there's an asymmetry. The Internet is an order of magnitude larger > than most other networks. > Well I strongly support the domain name system. I was just stirring in case there was some unconscious arrogance going on. Since it is conscious arrogance I'll just forget the whole thing. Bob Smart <smart@ditmela.oz.au>
dcrocker@DECWRL.DEC.COM (Dave Crocker) (10/25/89)
I hope that everyone takes note of your simple, but fundamental observation that use of '%' is a) a legal part of the local-part of an RFC822 address, and therefore b) its use is strictly up to the administrator(s) of the host that interprets the local-part. I.e., the host referenced in the right-hand-side of the address. The Host Requirements working group spent quite a bit of time considering whether to make the %-hack a formal part of the document. We decided that such a section would formally constrain something that is, by definition, a matter of local choice. (I suppose we could travel down the road of analogies, looking at state vs. federal rights, here, but I can't think of a funny ending to it.) Use of MX requires formal, inter-organization registration and maintenance. Use of the %-hack is a much more localized decision and may be modified much more easily. The first makes Internet email users think that you are actually on the Internet, the latter means that new off-net mail hosts do not need to be registered in the Domain system. Dave
vjs@rhyolite.wpd.sgi.com (Vernon Schryver) (10/26/89)
There was a time when I sneered at the %-laden From: lines from berkeley.edu and other places. That time has passed. Sgi.sgi.com now rewrites From:'s for internally originated mail to something of the form user%host.dom.sgi.com@sgi.sgi.com. (Of course, stuff passing thru is not touched.) This happened when people exhausted my tolerance to complaints that replies to our mail did not work. It took hundreds of such complaints before I yielded. Our problem was made worse by many of things, including my mistakes, broken DNS-servers out there, having thousands of hosts in several domains behind the sgi.sgi.com gateway, not wanting to burden our secondary servers with our not too small and ever changing host files, and the security worries of some people. We tried using simple wildcard MX records. That does not work. The obnoxious "%" does work. Vernon Schryver Silicon Graphics vjs@sgi.com
randall@uvaarpa.virginia.edu (Randall Atkinson) (10/26/89)
In article <43522@sgi.sgi.com> vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes: >Our problem was made worse by many of things, including my mistakes, broken >DNS-servers out there, having thousands of hosts in several domains behind >the sgi.sgi.com gateway, not wanting to burden our secondary servers with >our not too small and ever changing host files, and the security worries of >some people. >We tried using simple wildcard MX records. That does not work. The >obnoxious "%" does work. Actually if you'll check the GE nameserver, you will find that we have hidden our entire internal DECnet (hordes of ever changing hosts) behind an MX record for *.DNET.GE.COM and the domain I look after is has a MX for *.CHO.GE.COM. We had a *&^% of a time when we tried to observe the "%" hack and I have had _zero_ complaints from users since we got our MX records straight last spring and eliminated all usage of the "% hack" whereas I had near daily gripes until the changeover about lost or delayed mail. We exchange mail regularly with most of the known Internet including Asia and Europe. No problems. I wonder if SGI's problem was really getting the MXs straight rather than something else. Use of the %-hack can hide from view problems with one's nameserver's MX records. My own experience is why I feel so strongly about eliminating the %-hack from common usage and reserving it for the few special cases wher an MX record won't work for political or technical reasons and no other solution is present. I hear gripes that MILNET users are the problem since many don't have nameservers. I haven't experienced any problems with folks on MILNET myself though. Ran GE-Fanuc North-America
brian@ucsd.Edu (Brian Kantor) (10/26/89)
UCSD continues to rewrite outgoing Internet From: lines as user%campushost@ucsd.edu because that is the only format that works with everyone we correspond with, especially &^*%^&*$ MILNET hosts that are still using the damn hosts table and will be until the next war. And I refuse to register all 2000+ of our hosts with the NIC (no doubt they're overjoyed by that!). We have MX registrations for all our campus hosts that can receive mail, and we'll cheerfully accept mail for them in the form user@campushost.ucsd.edu but we aren't going to show that as a return address until I'm well convinced that I'm not gonna have screaming faculty members phoning me up to tell me about some babyburner on a FOONLY somewhere on MILNET whose permanent latrine orderlyXXXX er, system manager says they can't send us mail because our hostname isn't in their hosts file. I've been down that road before. Warmest personal regards, - Brian
pcg@rupert.cs.aber.ac.uk (Piercarlo Grandi) (10/27/89)
In article <31038@news.Think.COM> barmar@kulla (Barry Margolin) writes:
1) Users have to remember gateways. smart@ditmela.oz.au is easier to
remember than smart%ditmela.oz.au@uunet.uu.net.
2) % isn't the only such hack, and it's not always clear what happens when
multiple hacks are used together. What is the meaning of foo!bar%baz@quux?
Is it quux -> foo -> baz -> user bar or quux -> baz -> foo -> user bar?
True, true, and this is why domain names are *a good thing*.
3) Problems occur when network topology changes. If uunet.uu.net is
replaced as the Internet->UUCP gateway, all the %host@uunet.uu.net
addresses are invalidated.
Ahhhhh. This is the crux: %host@uunet.uu.net is a *route*, not an address.
In the absence of centralized administrative control, users will *always*
be required to eventually use routes; the address->route translation cannot
be *always* performed by some kind of distributed database, without centralized
administrative control of *both* the naming and transport.
I am one of those that think that centralized administrative control is not
only impossible, it is also not very desirable. People that do not see beyond
the Internet think otherwise on both accounts.
This means that users must learn a new gateway,
and mailing lists all over the place have to be updated. The domain system
simply requires updating the database, and users never notice.
Ahhhhh. Another difference of point of view. You are assuming
there is a benevolent entity that automagically updates in real
time (or close enough) the naming and routing distributed
databases. This does not happen in the real world, *even* on the
Internet (where it is close enough though). Lazy or untrained
system administrators, mistakes, etc...
5) You're dependent on an arcane, nonstandard feature being implemented on
a host over which you have no control. Uunet may decide at any time to
stop supporting %, and you'll run into problems.
Besides encouraging hosts to switch away from %, I think they are also
encouraging new systems NOT to implement it in the first place.
Yes, but what is the alternative for source routing? The multiple
"@" hack is not an answer... Note that you cannot say "don't do
it", unless you have centralized control over naming and
transport.
Again and again I am skeptical of the feasibility of doing this;
it is already difficult (and I occasionally think it is
impossible) to switch away from relative naming, because it is
difficult enough to have centralized control over the namespace;
to have centralized control over transport is probably impossible
or even undesirable.
Again and again I think (nostalgic) that if everybody had adopted
the Usenet/uucp bang notation for BOTH naming and routing
everybody would be happier now (well, I would have liked to have
more then 7 chars for host names, to make it easier to have
unique names, or to support dot notation for domains).
More importantly, it sends a signal to implementors not to try related
kludges. For instance, I've heard of systems that try to parse non-local
local parts, in an attempt to optimize routes. They see foo%bar@baz,
realize that they know a host named "bar", so send the message directly to
bar rather than routing through baz; this loses if baz doesn't interpret %
as they think it does, or if baz has a different notion of the host bar.
Amen! May the heathen be stricken by data rot!
--
Piercarlo "Peter" Grandi | ARPA: pcg%cs.aber.ac.uk@nsfnet-relay.ac.uk
Dept of CS, UCW Aberystwyth | UUCP: ...!mcvax!ukc!aber-cs!pcg
Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg@cs.aber.ac.uk
barmar@kulla (Barry Margolin) (10/29/89)
In article <PCG.89Oct27134552@rupert.cs.aber.ac.uk] pcg@rupert.cs.aber.ac.uk (Piercarlo Grandi) writes: ]In article <31038@news.Think.COM] barmar@kulla (Barry Margolin) writes: ] This means that users must learn a new gateway, ] and mailing lists all over the place have to be updated. The domain system ] simply requires updating the database, and users never notice. ] ]Ahhhhh. Another difference of point of view. You are assuming ]there is a benevolent entity that automagically updates in real ]time (or close enough) the naming and routing distributed ]databases. This does not happen in the real world, *even* on the ]Internet (where it is close enough though). Lazy or untrained ]system administrators, mistakes, etc... Oh, and who is the benevolent entity that fixes all the explicit routes in mailing lists, not to mention the ones in people's heads? My assumption is that the MX information is stored in fewer places, so fewer benevolent entities need be involved. It's true that if a major forwarder, such as uunet, goes away the necessary updates are extensive in both cases, but I think they're orders of magnitudes more extensive when explicit routes are involved. When a fringe forwarder goes away it only affects a small portion of the domain database, but the affected hosts may be referenced in files on many other hosts. Barry Margolin, Thinking Machines Corp. barmar@think.com {uunet,harvard}!think!barmar