[ih] Some BITNET history, was bang paths

[Marty Lyons]

On Jan 21, 2010, at 1:51 PM, Ofer Inbar wrote:

> I do remember using hacks like @sitename.bitnet on non-BITNET sendmail
> hosts that were configured to rewrite those to go to the appropriate
> gateways.


The .bitnet mail suffix appeared first at the BITNET relay points, including Wisconsin and Princeton; I think that may have been around 1985.   On the BITNET side, many people were running the excellent MAILER MTA written by E. Alan Crosswell who was at Columbia University.   MAILER allowed sites to send files out on a default RSCS connection or an IP path if they had an interface.   At the time, connecting an IBM mainframe channel to Ethernet was a big deal (more below).

Mail on an IBM VM system had no special status as such.   Mail files appeared in the user machine's virtual reader, and were sent to the virtual punch.   The Remote Spooling Communication Subsystem (RSCS) could enqueue files to remote hosts and a specified virtual machine.   For example, our node was NJITCCCC (NJ Inst. of Tech.).   A file could be sent from another node/user pair to NJITCCCC MARTY, with no expectation or need for RFC compliant headers, or any "mail" related information -- it was just a file.   When MAILER became available, all files would get routed to NJITCCCC MAILER, which expected valid mail headers etc, and then would drop the validated file into the individual user's virtual reader.

One way to view the difference between the networks was in the ARPANET everything looked like a packet.  In BITNET, everything looked like a file.

The included MUA on IBM VM systems was the NOTE program, which only supported RSCS node names.   There was no way using just NOTE to send mail to the Internet at large as due to lack of headers mail would be dropped; intra-BITNET would work fine.   Richard Schafer at Rice University wrote the MAIL program which built valid RFC compatible mail files, and combined with MAILER, really allowed BITNET hosts to function as first-class Internet mail citizens.   Mail gatewaying to the Internet at large would transit through someplace like Wisconsin, Princeton, etc, and mail headers rewritten using the % routing hack as MARTY%NJITCCCC.BITNET at puccvm.princeton.edu.   Sometimes you'd see a MARTY at NJITCCCC.BITNET in the wild, but you'd have to hope your local postmaster had the MTA configured to get it back to the right gateway.   It was still an era when users benefited from local knowledge of how the underlying transport worked, to get things where they needed to go.

To handle the emergence of mailing lists, Eric Thomas wrote the second generation of mailing list server, LISTSERV, superseding the first version run at City University of New York which could not handle the load.  Ultimately Eric formed L-SOFT to market and sell LISTSERV, and it remains one of the leading commercial mailing list tools.

Fairly quickly the central nodes on BITNET started overloading, the more central nodes (e.g. CUNY, PRINCETON, WISCVM, CORNELL) starting manually managing RSCS mail queues.   CUNY in particular had a dedicated person routing files by hand as parts of the network would slow down.   To make matters worse, RSCS allowed instant messages using the MSG/TELL command (e.g. "TELL ARTY AT CUNYVM Want to get dinner?").   Originally these messages were designed for high-priority traffic such as operator messages to users, and in the network took precedence over mail traffic.   

Then Jeff Kell at Univ. of Tennessee wrote RELAY (probably the first distributed chat service) [1], and you can imagine what happened next.    Within a year there were restrictions on time of day when people would turn on RELAY (usually only late at night) otherwise the network would grind to a halt.  Most sites were connected using 9600bps leased lines...that was for a whole university -- something to keep in mind now that we all complain about our personal megabit connections :)

To connect to an Ethernet, IBM's official hardware offering included a 7170 DACU (Direct Attach Channel Unit) which consisted of a modified PC and block channel card; it was a poorly supported project and did not catch on.   Around 1988, IBM shipped the 8232 LAN Channel Station, which was a PS/2 with a microchannel card to connect to a mainframe channel (price was about $40k USD); Jeff Kravitz [2] worked on that project.  An independent company, Bus-Tech, Inc. ("BTI") sold their own interface box, which was both faster and cheaper than the IBM 8232.  The only problem was their driver didn't work well.  David Lippke from the University of Texas at Dallas wrote one which became the standard ("the Lippke driver") which was recommended by BTI.   

IBM during the early days of BITNET didn't offer a TCP suite.   Independently, the University of Wisconsin wrote a full package known as WISCNET (around 1985), which IBM later sold as a supported product.    Matt Korn wrote the SMTP component while at Wisconsin, and later moved to IBM.   When the product became supported within IBM, it was handled at TJ Watson Research Center; Jay Elinksy and Barry Appelman worked with Matt on the product.

Once VM systems has full IP access, other utilities rapidly were developed; Peter DiCamillo at Brown University wrote tn3270, to allow telnet in 3270 full screen emulation mode.   I seem to recall the first version was developed as a Mac application, then a Unix port was made that used the curses library.  Arty Ecock at the City University of New York wrote RXSOCKET for socket support.  Rick Troth at Rice University wrote a Gopher implementation.

Many of the significant contributors to BITNET attended VM Workshops, held in the summer at college campuses (at low cost), run by Harry Williams from Marist College.   The IBM SHARE VM User Group also was a big event.    The BITNET and VM community was small and close knit, and seemed to succeed despite IBM sometimes (who continued to push their MVS operating system as a superior alternative to VM... even though it wasn't!).

Several people from IBM had moved to Advanced Networks & Services (ANS) when it was formed in the late 80s. (An aside:  In 1993, I joined Matt, Jay, Barry, and David in the early days of America Online.   Our first IP connections were from ANS.  UUnet, MCI, and MAE-EAST were just down the street; SURAnet was based at the University of Maryland.   The DC area was a real hot spot for the net, to some degree even more than the valley.)

The underlying BITNET network remained a flat namespace due to RSCS being the routing layer.   BITNET never moved to hierarchal names, although multi-homed hosts might appear that way on the ARPA side -- CORNELLC was a big BITNET node and was also cornellc.cit.cornell.edu.    Princeton's main node was PUCCVM aka puccvm.princeton.edu.   CUNYVM was cunyvm.cuny.edu.   In the later years (1988) BITNET2 allowed nodes to route traffic over the Internet (RJE over IP).    With the emergence of the regional networks, within several years however most campuses had dedicated Internet connections and BITNET registrations started declining; BITNET effectively wrapped up by 1996.

The interesting trend in nearly all of Internet history is the same people keep crossing paths repeatedly; I've lost track of how many people I keep bumping into over and over just with a different business card.   And the overlap between operating system development and network development is significant -- more so than has been written in the popular press, although John Day covers this very well in his book [3].    

Many of the BITNET milestones were enabled by the VM operating system, which was a great platform for development and testing.   Ultimately even IBM realized that the network and how easy it was to utilize was critical to sales; there was a long held belief within IBM that SNA (their proprietary architecture) and 3725 style controllers were the future of networking [this was often attributed to Ellen Hancock].   The dramatic sales of low cost DEC VAX systems in the mid-80s really showed how important networking capabilities were "out of the box".  Although oddly enough, VMS in the 80s did not natively support TCP, as Digital was pushing DECNET.   One of the best implementations under VMS was from a company called TGV ("Three Guys and a Vax") in Silicon Valley.  

So in the case of BITNET, much of the innovation was driven by the users despite an operating system and hardware that really didn't offer much in the way of native networking functionality.  From hardware interfaces to most of the software, a great deal of the technology was developed outside of the organization that built the underlying systems.  Much of this was out of necessity since BITNET was the only game in town for many campuses, since it was so difficult to get approval for a full ARPAnet connection.

Marty

[1] http://en.wikipedia.org/wiki/Bitnet_Relay
[2] http://domino.research.ibm.com/comm/research_people.nsf/pages/jkravitz.index.html
[3] "Patterns in Network Architecture", Prentice Hall, 2008, ISBN 978-0132252423