[ih] bang paths, was Domain Names

[Marty Lyons]

On Jan 20, 2010, at 2:55 PM, Dave CROCKER wrote:

> On 1/20/2010 8:37 AM, Craig Partridge wrote:
>> The assimilation of USENET (and CSNET and to a lesser degree BITNET) into
>> ARPANET was intentionally enabled...
> 
> Each of these had their own syntax.


One of the challenging things in the 1980s was getting mail from one network to another.   Figuring out how to manually route through gateways was something of a black art -- and often not officially sanctioned... as mail loads got heavier, sometimes postmasters would ask for people to stop using their connections.   

At the time, getting a connection to ARPAnet was still quite a non-trivial process; apart from the large research universities very few places had a circuit.   CSNET and BITNET both helped fill that gap since NSF approval wasn't required.  In the case of BITNET all a site needed was an IBM system running VM (or a VAX which could speak the IBM RSCS protocol), and a friendly partner who would accept a leased line connection.   The ARPAnet had already solved the HOSTS.TXT problem by moving to DNS, but BITNET maintained a flat file of all connected hosts (and their peers) which was updated manually; routing was handled by RSCS, no explicit path knowledge was needed by the sender or appeared in the mail header (user at node sufficed).  UUCP relied on explicit path routing ("bang paths": e.g. !rutgers!decwrl!user) until pathalias [1] became widespread and addresses of the form user at node.uucp were supported.

There was a mailing list dedicated to network topics known as INFO-NETS hosted at MIT.    After about a year of research I had a fairly complete map showing all the mail gateways.   As a point of historical interest in 1991 I sent a copy to John Quarterman and it appeared in Matrix News (Volume 1, Number 6; September 1991).

The "well known" gateways were also well-connected and on multiple networks.   Typically they had mailers which could reliably parse addresses, rewrite headers as needed, and get mail moving on to the next hop.    Manually routed mail could use the percent routing hack, quoted components, source routing, or weird combinations of all of them at once.    It was really pretty amazing some of the stuff that would work.  Lots of people in that era became sendmail wizards (often against their will!).  On any given network, people knew the host of last resort by heart... you'd hear lots of "just send it to (ucbvax, wiscvm, ihnp4), they can probably get it there".

A whacky example: Mail from a MAILNET connected host to someone inside DEC (routed via their UUCP connection):   
"rutgers!decvax!decwrl!KYOA::USER"%rutgers.arpa at mit-multics.mailnet

Some well-connected sites included:  Cal Berkeley (UCBVAX, UCBARPA, BERKELEY) [ARPA, UUCP], Univ. Delaware (UDEL-RELAY) [ARPA, CSNET], Univ. of Wisconsin (WISCVM) [ARPA, BITNET], DEC Western Research Lab (DECWRL, DECVAX) [ARPA, CSNET, UUCP], AT&T Bell Labs at Naperville, IL (ihnp4.uucp) [UUCP], Rutgers (RUTGERS) [ARPA, UUCP], 

Up until the early 1990s, it was still difficult to get on "the network".   Many sites were still relying on the simplest way to move mail, which was UUCP.  It wasn't until the NSF Acceptable Use Policy was lifted and the regional and independent networks were available (e.g. PSInet), that moving packets became straightforward.

On a related topic: People had lots of fun on internal networks coming up with host naming schemes.   It was one of the best parts about being the systems person.  I recall MIT has breakfast cereals for awhile (MIT-RICECHEX).  Rutgers main machines were colors (RU-RED, RU-BLUE, RU-GREEN).  When I set up the namespace at South Pole Station, all the hosts were named after the one thing you'd never see: trees.

A great reference on the history of various networks and how they all worked together is John Quarterman's "The Matrix: Computer Networks and Conferencing Systems Worldwide", Digital Press, 1996, ISBN 1555580335.  I'm not sure if John is on the list.

Some more mail gateways from 1985:

ARPA-UUCP: brl-tgr, ucbvax, berkeley, seismo
ARPA-ACSNET (Australia): seismo.arpa
ARPA-AUGMENT (Xerox): office.arpa
ARPA-Xerox: xerox.arpa
ARPA-CSNET: csnet-relay.arpa, csnet-sh.arpa, csnet-cic.arpa
ARPA-BITNET: wiscvm.bitnet, ucbvax.arpa
ARPA-DEC (Digital Equip. Corp.): decwrl.arpa
ARPA-CCNET (Columbia Univ): columbia.arpa, cu20xx.arpa
ARPA-MAILNET: mit-multics.arpa
ARPA-JANET (U.K.): ucl-cs.arpa
CSNET-UUCP: tektronix.csnet, tektronix.uucp
CSNET-DFN (Univ. Karlshruhe, Germany): germany.csnet
CSNET-XEROX: xerox.csnet
CSNET-HUJI (Herbrew Univ., Jerusalum, Israel): israel.csnet
CSNET-SDN (Seoul, Korea): kaist.csnet
CSNET-MTSNET (Univ. of Michigan): umich.csnet
CSNET-DEC (Digital Equip. Corp.): decwrl.csnet
CSNET-EAN (Canada): ubc.csnet
CSNET-SUNET (Swedish Univ. Network): CHALMERS.CSNET
CSNET-JANET (U.K.): ucl-cs.csnet
BITNET-UUCP: psuvax1.bitnet, psuvax1.uucp
BITNET-CSNET: wiscvm.bitnet
BITNET-IBM (IBM Research): vnet.bitnet, cunyvm.bitnet
BITNET-NETNORTH (Canada): canada01.bitnet
BITNET-CCNET (Columbia Univ.): cuvma.bitnet
BITNET-EARN (European Academic Research Network, via GSI Darmstadt): dearn.bitnet
MAILNET-JANET (U.K.): ucl-cs-mailnet.mailnet
MAILNET-CCNET (Columbia Univ.): carnegie.mailnet
MAILNET-MTSNET (Univ. of Michigan): umich-mts.mailnet
JANET (U.K.)-PSS (U.K.-British Telecom PDN): pssa.pss
JANET (U.K.)-UUCP: dcl-cs.uucp, !seismo!mcvax!ukc!icdoc!dcl-cs!davis
ACSNET (Australia)-UUCP: !decvax!mulga!psych.uq.oz
DEC-UUCP: decwrl.uucp

[1] http://www.uucp.org/papers/pathalias.pdf