Network Working Group                           A. McKenzie
RFC 241                                       BBN
NIC # 7671                                      29 September 1971
Categories: B.1, C.1, I.1
Updates: none
Obsoletes:  Our Previous Verbal Comments

                   CONNECTING COMPUTERS TO MLC PORTS
                   ---------------------------------

        Several times we have been asked if computers can be con- nected
   through serial communication lines to ports on the Terminal IMP's
   Multi-Line Controller (MLC) [related questions about the level of
   software support provided by the Terminal IMP to such a connection,
   have also been raised].  In the past we have said, "Please don't!" We
   now say, "Sure, but will that really help you the way you think it
   will?"

        (1) Connections between computers and IMPs (i.e., the Host
   interfaces) have been assumed to be error-free.  This assumption is
   justifiable on the basis that the IMP and Host computers were
   expected to be either in the same room (up to 30 feet of cable) or,
   via the Distant Host option, within 2000 feet on well- controlled,
   shielded cables.  A connection through common carrier facilities is
   not comparably free of errors.  Usage of common- carrier lines for
   connecting a terminal to an IMP, including the assumption of a human
   at the terminal, is a situation in which the typical errors which do
   occur can be accommodated.  Usage of the same wire, with the same
   typical errors, for a computer-to- computer connection is likely to
   be a situation in which the errors are unacceptable.  The present
   version of the Terminal IMP does not provide error control either
   within its hardware or within its software on any ports of the
   Multi-Line Controller.  Further, we feel that computer-to-computer
   connections over common carrier circuits should employ strong error
   control, such as that

                                                                [Page 1]

RFC 241

   used on the IMP/IMP circuits, and that attempts to use minimal error
   control (e.g., character parity) is an undesirable technical choice.
   Strong error control, with its retransmission scheme, not only would
   imply significant changes in the Terminal IMP, but a non-trivial
   hardware/software implementation at the remote computer end of the
   circuit.

        (2) Because the Terminal IMP has many obligations, the share of
   its bandwidth which can be given to a Host coming in over the MLC
   will be small.

        (3) The command language provided at a port of the Multi- Line
   Controller was designed with terminals and people in mind.  It
   provides very few of the capabilities which a computer requires in
   order to effectively utilize the communication network.  For example,
   only a single pair of connections can be made from a given Terminal
   TMP port; Host computers generally desire a larger number of
   simultaneous connections to other Hosts on the network.  Assuming the
   present Host/Host protocols, such a Host could not conveniently act
   as a server.

        If, despite these potential difficulties, connection of a
   computer to the network through an MLC port appears to be useful, BBN
   has no objection.  In fact, we would be extremely interested in
   hearing about actual experience with this type of network connection.

   AMcK:jm

         [ This RFC was put into machine readable form for entry ]
         [ into the online RFC archives by BBN Corp. under the   ]
         [ direction of Alex McKenzie.                   12/96   ]

                                                                [Page 2]