rfc88.txt

来自「RFC 的详细文档！」· 文本 代码 · 共 508 行 · 第 1/2 页

Network Working Group                                          R. Braden
Request for Comments: 88                                        S. Wolfe
NIC: 5668                                                       UCLA/CCN
                                                         13 January 1971


          NETRJS - A THIRD LEVEL PROTOCOL FOR REMOTE JOB ENTRY

   A.  Introduction

   NETRJS is the name for a message protocol and set of control
   conventions which will allow users at remote Hosts to access the RJS
   ("Remote Job Service") remote batch subsystem of CCN.  RJS[1] was
   written at CCN to support remote batch (car reader/line printer)
   terminals over communications lines.

   RJS makes a remote batch terminal's unit record devices operate as if
   they were at the central site; thus, a remote user enters OS/360
   jobs, complete with JCL, into the remote reader.  The jobs are
   spooled into the operating system and run in their turn, and the
   printed and/or punched output is returned to the remote terminal from
   which the jobs originated (unless the user or operator re-routes the
   output).  The remote terminal may also include a console typewriter
   to be used by the remote operator to receive and send messages and to
   exert control over his terminal [2].

   When RJS is used via the ARPA Network, the "remote terminal" is
   expected to be a multiprogrammed user process in a remote Host.  We
   will use the RJS term "remote site" for such a user process, which
   presumably simulates unit record devices by file I/O.  Furthermore,
   several users at the same remote Host may simultaneously use NETRJS,
   acting as independent "remote sites" distinguished by 8-character
   names called _terminal-ids_ (because each remote site appears to RJS
   as a separate physical terminal).  Valid terminal-ids will be
   assigned to individual users or user groups at remote Hosts who wish
   to use NETRJS.

   Under NETRJS, a separate ARPA network connection is opened from this
   remote site to CCN for each (simulated) unit record device.  Each
   such connection will be called a _channel_ and be designated _input_
   or _output_ with reference to CCN.  We define a _standard_ remote
   site in NETRJS to have the following five channels (See Figure 1):

      1._Operator Input Channel_ - Commands and messages entered by
         remote "operator" console.

      2 _Operator Output Channel_ - Message stream which would normally
         be directed to remote operator.



Braden, et. al.                                                 [Page 1]

RFC 88              NETRJS - A THIRD LEVEL PROTOCOL      13 January 1971


      3._Input Stream_ - One simulated Hollerith card reader for job
         submission.

      4._Printer Stream_ - One simulated line printer to record printed
         output (system messages and SYSOUT data sets) from jobs.

      5._Punch Stream_ - One simulated card punch, capable of recording
         arbitrary (i.e., transparent) binary text.

   RJS actually will support more than one reader, printer, and punch at
   each remote terminal, so the NETRJS protocol could easily be expanded
   to allow multiple simultaneous I/O streams to each Network user.
   However, this does not presently appear useful, as the ARPA Network
   bandwidth will normally be the limitation on the transmission speed
   under NETRJS.

   Under NETRJS, the text of a single network message is called a
   _block_.  A block is of variable length, up to 900 bytes (except
   operator input and output blocks, which may not exceed 130 bytes).
   Here the term _byte_ refers to the set of 8 bits representing one
   character; each byte is to be aligned on an 8-bit boundary within the
   message (and block).  Thus we may consider a block to be a string of
   bytes.  The detailed format of a block will be defined in Sections E,
   F, and G, using essentially the formalism suggested by Bobrow and
   Sutherland in RFC #31.

   Since the central site Host (CCN) is an IBM 360, NETRJS uses the IBM
   EBCDIC character code to avoid redundant code conversion at both
   hosts in those cases when the remote host also uses EBCDIC
   internally.  However, the message formats make no assumption about
   the code, and in fact, "object decks" sent to the (simulated) card
   punch will normally contain arbitrary binary text.

   To maximize the use of the available Network bandwidth, we strongly
   recommend transmitting input blocks as large as possible; CCN will
   always fully block NETRJS output.  Furthermore, to avoid excessive
   overhead, we urge that all NETRJS users make their marking _a
   multiple of 8 bits_, so the messages received at CCN arrive on a byte
   boundary.

   B.  Starting a Session[3]

   The initial connection protocol for NETRJS is essentially that of
   Crocker in RFC #66 (as restated by Harslem and Heafner in RFC #80),
   with some extensions.  User U at a remote Host presumably requests
   his outgoing logger to make a NETRJS connection to CCN.  This





Braden, et. al.                                                 [Page 2]

RFC 88              NETRJS - A THIRD LEVEL PROTOCOL      13 January 1971


   logger does so by first sending an initial RFC to connect socket
   (user,aen) = (U,s) to CCN socket (0,5).  User 0 at CCN is the
   incoming logger, and aen = 5 signifies NETRJS.

   The CCN incoming logger will allocate a set of (six) consecutive aen
   numbers A, A+1,......A+5, for user U, return a message containing the
   socket number (U,A) as specified in RFC #66, and close the initial
   connection.  The remote and central sites will then open an input
   channel between CCN socket (U,A) (socket f in Figure 1) and remote
   socket (U, s+1).  This is the remote operator input channel.  The
   other devices have fixed aen's at CCN assigned relative to A, in
   particular:

                                 CCN Socket
         Channel                 (User,aen)

         Operator Input          (U,A)
         Operator Output         (U,A+1)
         Card Reader (No. 1)     (U,A+2)
         Printer (No. 1)         (U,A+3)
         Punch (No. 1)           (U,A+5)

   Once the operator input channel is open, the remote site must
   transmit a valid RJS signon message [2].  This message is free-format
   and consists of the command verb "SIGNON" followed by the user's
   terminal-id.  If RJS does not recognize the terminal-id or has no
   available Line Handler for the Network, it will indicate refusal by
   closing the operator input channel.  Central site issues subsequent
   RFC's for the other channels listed above only in response to
   corresponding RFC's from the remote site

   To terminate the session, the remote site may close the console input
   channel (socket "a" in figure 1).  Alternatively, the user can enter
   a SIGNOFF command through the operator input channel; in this case,
   RJS will wait until the current job output streams are complete and
   then terminate the session.  RJS terminates the session by closing
   the console output channel (socket g).  Also, if RJS should abend
   then socket g will close.  If either site terminates the session, all
   other connections for this remote site should be closed.  Note that a
   user can submit a number of jobs, sign off, and later receive his
   output when he signs on again.

C.  Channel Control

   Flow control in NETRJS is handled by the Network protocol ALL
   mechanism.  Before transmission of a stream of records can begin on a
   particular channel, the remote site must issue an RFC and Central
   must reply.  This allows the central site to determine the remote



Braden, et. al.                                                 [Page 3]

RFC 88              NETRJS - A THIRD LEVEL PROTOCOL      13 January 1971


   configuration dynamically.  A particular card reader, printer, or
   punch channel is open only while it is active, so the receiver need
   not tie up buffer space needlessly.  Each of these channels, when
   open, assumes a buffer allocation of at least 900 bytes at the
   receiver.

   The operator input and output channels, on the other hand, are open
   throughout the session.  On these channels the receiver must provide
   an allocation of at least 130 bytes.

   After sending the SIGNON command over the operator input channel, the
   remote site should send RFC's for all output channels which are ready
   to receive data.  When output is available for that site, Central
   returns an RFC and begins transmission.  Central closes an output
   channel (socket i and j) at the end of the output for each complete
   batch job.[4]  The remote site must then send a new RFC and Central
   must reply with an RFC to start output for another job to that
   device.  This gives the remote site a chance to allocate a new file
   for each job without breaking the output within a job.  If the user
   at the remote site wants to cancel (or backspace or defer) the output
   of a particular job, he enters appropriate RJS commands[2] on the
   operator input channel.

   When the remote site is ready to submit a job (or stack of
   consecutive jobs), it issues an RFC for the card reader input
   channel.  The remote site is not required to close the channel
   (socket c) after each job in a "stack" of jobs, but he must close it
   following the last job in the stack to initiate its processing.

   It may be necessary for the receiver site to abort a particular
   channel, perhaps due to a transmission error (see Section D below on
   checking) or a disk I/O error.  The receiver may abort a channel
   (other than console output) by closing it (sockets d, e, f, and h).
   This action signals the transmitter to re-transmit the information
   after the channel has been reopened (initiated by the remote site, as
   always).  The transmitter, on the other hand, aborts a channel by
   sending a block with a particular bit combination (e = 2 in BCBYTE;
   see Section E).

   If either site aborts card reader (input) channel, RJS will discard
   the text of the last partially-spooled job; the remote site should
   re-transmit this job.  Note that repeating an entire stack will enter
   duplicate jobs into the system, but the second copy of a job will
   "flush" due to its duplicate job name.

   If a printer or punch (output) channel is aborted, Central will re-
   transmit from the beginning of the current SYSOUT data set; the
   effect is the same as a RESTART command.[2]



Braden, et. al.                                                 [Page 4]

RFC 88              NETRJS - A THIRD LEVEL PROTOCOL      13 January 1971


   If the operator input channel is aborted, the remote site must re-
   transmit the last _block_.  Finally, the operator output channel has
   no abort condition defined.  Central will never send Channel Abort
   message on this channel; if the remote site closes its socket (socket
   b), Central will not re-transmit, but simply cease sending messages
   until the channel is reopened.  Therefore a remote site can operate
   without an operator output channel; however we do not recommend this,
   as the user will then miss operator advisory messages such as a
   warning of an impending IPL.

D.  Checking

   The nature of remote job entry service is such that a low rate of
   undetected errors is mandatory.  The IMP's use CRC's and sequence
   numbers over the communication lines, so the effective IMP-IMP error
   rate should be negligible.  Although there is no checking provided
   for the IMP-Host interface, it seems likely that these interfaces
   will either be reliable or fail catastrophically; it seems unlikely
   that "drop-outs" or other random failures will occur.  Therefore only
   the following simple checks are provided:

   1. Each block will (at least initially) contain a fixed bit check
      pattern using both on and off states of each bit path in the 16
      bit PDA interface at CCN.