Network Working Group                                         A. Bhushan
Request for Comments: 327                                        MIT-MAC
NIC: 9261                                                 April 27, 1972


                 DATA AND FILE TRANSFER WORKSHOP NOTES
	
   On April 14 and 15, 1972, a Data and File Transfer Workshop was held
   at M.I.T., Cambridge, Mass.  A list of attendees of the meeting for
   April 14 and 15 is appended to the notes.  This note attempts to
   summarize most of the topics discussed and all of the decisions
   reached at the workshop meeting.

   The following is a summary of the talks and discussions on April 14,
   1972.

   Steve Crooker discussed a general theory for Network protocols.
   Protocols transformations should have a unique inverse, and should be
   transitive.  Transformation to a standard form requires only 2n
   transformations (n = number of different types of hosts), as compared
   with n(n-1) transformations with no standard form.  A standard
   approach is preferable for n >= 3.

   For file transfer, one could define a Network Virtual File Image.
   There was some discussion on whether it was possible to satisfy the
   above rules for file structure transformations.  No agreement was
   reached and the problem was abandoned for the present.

   Further discussion lead to the following formulation of the Workshop
   goals:

   To come up with data and file transfer protocol/strategy that
   satisfies the needs of ARPANET users including Maxi-HOSTs, Mini-
   HOSTs, TIPs, Datacomputer, RJE, and Mailbox users.

      Goals for the protocols/strategy were set as:

      1.  It should preserve the integrity of data.

      2.  It should preserve the integrity of character representation
          and interpretation.

      3.  It should preserve the integrity of structural information, to
          the extent conveniently possible.

      4.  It should lead to the development of a Network Virtual File
          System.

   Richard Winter discussed the Datacomputer application.  The
   Datacomputer though usable from terminals directly will not be
   engineered for direct terminal users, but for use by programs.  In
   Datalanguage a user can define data and file structure, and also how
   the file/data is to be transferred.  Using the data language it is
   possible to transfer entire files, or only the relevant parts of
   files.  The following is an example of file transfer as currently
   envisioned in the Datacomputer.

          LOGIN  <user> <password>
          CREATE  <file name> <description>
          CREATE  <port name> <description>
          PORT  <port name> <external name>
          <file name> = <port name> (for transfer to Datacomputer)
          <port name> = <file name> (for transfer from Datacomputer)
          LOGOUT

   (CREATE statements are needed only when the description(s) required
   are not already on file at the Datacomputer.  A port description can
   specify a standard "external name", thus making a port statement
   optional also.  "External name" is to be a HOST-socket specification.
   The data transfer is to be in accordance with network data transfer
   standards.  The File and Port descriptions are to be in
   Datalanguage.)

   Alex McKenzie discussed the TIP user needs, describing the current
   capabilities and limitations of TIPs and TIP terminals.  TELNET
   format is the first choice of TIP users, followed by DTP using the
   indefinite bit stream mode.  There are two TIPs with magnetic tape
   systems which are capable of transferring data between them using the
   current DTP (RFC 264) in the descriptor count mode (utilizing
   sequence number option).

   Bob Braden discussed the RJS protocol and presented some data on RJS
   use.  NETRJS is 1% of CCN job load representing 2,000 jobs, 10,000
   sessions and 1,000 hours connect time in the last 5 months.  Average
   job input is of the order of 100,000 bits (400 cards), average job
   output is 700,000 bits (1,000 lines).  Large files have about 10
   million bits representing about 8-10 minutes of transmission time.
   The RJS protocol will be defined in a forthcoming document.

   Ray Tomlinson described the CPYNET system BBN is using to transfer
   files among TENEX systems.  CPYNET commands are ASCII strings with a
   fixed syntax.  The original connection is closed after a command is
   accepted, and data is transferred on a new connection using previous
   socket number, but with possibly a different byte size.  The data
   transfer rate achieved in CPYNET has been about 10 Kb/s.

   Abhay Bhushan discussed the evaluation of network protocols and
   presented some preliminary measurement results.  The evaluation
   criteria for protocols should include speed (real time delay and
   transmission rate), efficiency (cpu time or cost), reliability (error
   rate and failure rate), convenience (ease of use and implementation),
   and usage (suitability for various application and user classes).

   The parameters that affect speed and efficiency for given system
   conditions (fixed load, etc.) are:

      1)  Byte size used for NCP connection.

      2)  Average message size used for transmission.

      3)  Data format conversion (e.g., into Network ASCII, DTP Blocks,
          etc.).

      4)  Buffer size and I/O mode used (unit or block mode, etc.).

      5)  Other protocol constraints (acknowledge, error checking,
          connection procedure, etc.).

   There was some discussion as to how data and file transfer protocols
   may be altered to make transfer faster and more efficient by using
   optimum byte size and minimizing some of the constraints that impose
   a large overhead.

   The follow up discussions on DTP and FTP lead to a list of discussion
   and decision items for the next day.  The following is a summary of
   decisions reached on Saturday, April 15, 1972.

      1.  Separate connections are to be used for data and control
          information.

      2.  Control connection is to be a "TELNET" full-duplex connection
          (NVT-ASCII), established via the ICP.  Data connections are to
          be simplex connections established directly.

      3.  The File Transfer and File System commands and their arguments
          shall be printable ASCII strings, instead of numeric codes, so
          that they are directly usable by a user at a terminal.  The
          interaction, however, will be optimized for usage by programs.
          (indirect use).

      4.  The byte size and user socket for data connection, data
          representation, and transfer mode to be used in file transfer
          may be chosen by a user via one or more commands requiring a
          positive or negative acknowledgment.

      5.  The following data representations are to be accepted by all
          servers:

          1)  Network ASCII (7-bit ASCII in 8-bit field with 8th bit
              zero).

          2)  Local Byte (a server option to store data in an efficient
              manner, the storage scheme should be well publicized).

          3)  Image (a sequence of bits which should be stored
              contiguously independent of the byte size chosen for
              transfer).

          4)  ASCII Print file (convert ASCII file to a form suitable
              for printing).

          5)  EBCDIC Print file (convert EBCDIC file to a form suitable
              for printing).

      6.  Record structures are allowed but not mandatory.  A user with
          no record structure in his file should be able to store or
          retrieve his file at any host.  If a serving host cannot
          accept record structure, it must inform the user of the fact.
          Any record structure information in the data stream may
          subsequently be discarded.

      7.  The following data transfer modes are defined:

          1)  Byte-Stream - End of File indicated by closing connection.
              No record structure.

          2)  Block - File is series of blocks preceded by a count
              field.  Appropriate means provided to indicate end-of-
              file, end-of-record, and restart markers.

          3)  ASCII - The file is network-ASCII, end-of-record, and
              end-of-file are indicated by a special "TELNET-control"
              character with 8th bit set to "one".

          4)  File is network-ASCII with an end-of-record defined by CR
              LF, and end-of-file by closing connection.

          5)  Hasp compressed file with end-of-record and end-of-file
              information.

      8.  A restart procedure will be provided to protect user from
          system failures (either host or process dying).  The issue of
          bits lost or scrambled is handled best at the NCP level.
          Standard error codes and responses will be provided for
          storage and I/O channel errors, at the FTP level.

          The restart procedure would require that the sender of data
          insert a special marker in the data stream (the marker has
          meaning only to the sender.  It could be bit-count, record
          count, or page count, etc.).  The receiver of data would mark
          the corresponding position of this marker in its own system,
          and return this information to the user.  In the event of a
          system failure, the user can restart the transfer by supplying
          this information with a restart command.

      9.  DTP is no longer a separate protocol but a set of transfer
          modes or format procedures whose use is defined by the file
          transfer protocol.

      10.  Abhay Bhushan will write the workshop notes and the draft
          specifications for the new file transfer protocol.

   LIST OF ATTENDEES, DATA AND FILE TRANSFER WORKSHOP

   Abhay Bhushan            MIT-MAC        April 14,15
   Bob Braden               UCLA-CCN       April 14,15
   Arvola Chan              MIT-MAC        April 14,15
   Steve Crocker            ARPA           April 14
   Eric Harslem             RAND           April 14
   John Heafner             RAND           April 14
   Chuck Holland            UCSD           April 14,15
   Alex McKenzie            BBN (NET)      April 14
   Bob Metcalfe             MIT-MAC        April 14
   Hal Murray               CCA            April 14
   Bill Plummer             BBN            April 14
   Jon Postel               UCLA           April 14
   Neal Ryan                MIT-MAC        April 14,15
   Marc Seriff              MIT-MAC        April 14,15
   Bob Thomas               BBN            April 14
   Ray Tomlinson            BBN            April 14
   Dick Watson              SRI-ARC        April 14,15
   Doug Wells               MIT-MAC        April 14
   Jim White                SRI-ARC        April 14,15
   Richard Winter           CCA            April 14,15

        [This RFC was put into machine readable form for entry]
     [into the online RFC archives by Hélène Morin, Viagénie 10/99]