Network Working Group                                      E. F. Harslem
Request for Comments: 141                                  J. F. Haefner
NIC 6726                                                            Rand
                                                           29 April 1971

             COMMENTS ON RFC 141 (A FILE TRANSFER PROTOCOL)

   1.  A file transfer protocol is needed.  Bushan's proposal would
   satisfy a particular current need that we have, as well as short-term
   envisioned needs.

   2.  Bushan's protocol would apear to be straight-forward in
   implementation, and extensible as claimed.

   3.  We would like to see implementations of such protocol be
   accomplished such that the file transfer program has general and
   complete access to the local file storage.  That is, it should be
   able to access a file that it did not create.  For example, if a
   program or user creates a file at site X (completely independent of
   the file transfer program), it would then be desirable to be able to
   retrieve the file via the file transfer program.  This is not a
   requirement of RFC #114 but we would like to see it implemented where
   possible.

   4.  Since implementation of a subset of transaction types is
   specifically permitted, we suggest inclusion of the following
   commands (in addition to append).

      insert records     within a file
      delete records     from within a file
      replace records    within a file

   Although these operations are not directly supported under IBM
   OS/360, we have used them with a non-standard file subsystem under
   IBM OS/360 and find them quite useful.

   5.  In addition to retrieve and lookup, get names of files under my
   access control would be useful.

   6.  The absence of status requests and responses is apparent.
   Although this is typically a function associated with a remote job
   entry (RJE) system, since the execute request is present it would
   seem appropriate to inquire about the status of the process created
   by the execute command.  This becomes increasingly more important
   where the execute is implemented as an RJE-like operation and
   scheduling time of the job might be prolonged.

   7.  When requesting execute, the using host sends parameters upon
   receipt of the rr response.  Executing a task can be implemented in
   several ways.  The options our 360 affords are RJE at job level and
   the attach macro.  Our preference would be the attach macro which
   immediately initiates an independent OS task within the partition of
   the program issuing the attach (presumably the File Service).  Such a
   task normally receives parameters upon initiation and can thereafter
   receive parameters from a program via some mechanism such as an event
   control block.  The second method requires special modifications to
   the program being executed; hence, it is not desirable.  Therefore,
   we either need the parameters included in the execute command or will
   not actually start execution until parameters are received.

   8.  Upon abnormal termination, one should include part or all of the
   spurious request as well as an identify- ing code to facilitate
   precise error recognition.

   9.  We would be interested in the outcome of the MIT/ Harvard
   experiments with the RFC #114 protocol.  What were the pitfalls,
   etc.?

         [ This RFC was put into machine readable form for entry ]
          [ into the online RFC archives by Simone Demmel 4/97 ]