Network Working Group                                         J. Kreznar
Request for Comments: 17                                             SDC
Category: Informational                                   27 August 1969

                  Some Questions Re: HOST-IMP Protocol

1.  Automatic deletion of links, as indicated in BBN 1822, page 11,
    seems bad:

     a) Link use may be dependent upon human use of a time share
        terminal - indefinite time between messages.

     b) Program using link may be slow due to:

        i)  Busy HOST (many jobs)

        ii) Much local I/O and/or CPU time between messages - is it
            that, if a HOST's user fails to use a link for 15 seconds,
            the HOST network program must generate a dummy message
            merely to keep the link open?

2.  Steve Crocker, HOST Software, 1969 Apr 7, asks on page 2: "Can a
    HOST, as opposed to its IMP, control RFNM's?"  BBN, Report No. 1837,
    1969 Jul, says on page 2: "The principal function of the (IMP)
    program...includes...generating of RFNM's..."  What if an IMP
    generates an RFNM and then discovers it cannot, for some reason,
    complete timely delivery of the last received message to its HOST?
    This seems especially pressing since I don't recall seeing anywhere an
    IMP constraint upon HOSTs that they must accept incoming messages
    within some specified maximum time.

3.  A HOST has to be prepared to repeat transmissions of a message
    into network (see, e.g., Page 17, BBN 1822) therefore why the
            special discardable NOP message (Page 12, BBN 1822).

4.  "Arbitrary delays," middle paragraph, page 23, BBN 1822, seems
    inconsistent with automatic link deletion questioned in 1 above.
    Normally the times involved differ by many orders of magnitude but a
    high priority non-network HOST responsibility could delay next bit for
    a long time.

    1.  Abhi Bhushan, Proj. MAC         10.  Sal Aranda, SDC
    2.  Steve Crocker, UCLA             11.  Jerry Cole,  "
    3.  Ron Stoughton, UCSB             12.  John Kreznar,"
    4.  Elmer Shapiro, SRI              13.  Dick Linde,  "
    5.  Steve Carr, Utah                14.  Bob Long,    "
    6.  John Haefner, RAND              15.  Reg Martin,  "

    7.  Paul Rovner, LL                 16.  Hal Sackman, "
    8.  Bob Kahn, BB & N                17.  C. Weissman, "
    9.  Larry Roberts, ARPA

         [ This RFC was put into machine readable form for entry ]
          [ into the online RFC archives by Marc Blanchett 3/00 ]

Network Working Group                                            R. Kahn
Request for Comments: 17a                    Bolt Beranek and Newman Inc
                                                             August 1969

                 Re: Some Questions Re: HOST-IMP Protocol

   THE FOLLOWING COMMENTS ARE IN RESPONSE TO JOHN KREZNAR'S QUESTIONS
   WHICH WERE RAISED IN NWG:- 17

   The deletion of a link entry from an IMP's link table will, in
   general, have no effect upon a Host transmission (or reception) at
   that IMP's site.  Let us distinguish between non-use of a link in-
   between messages and non-use of a link due to Host program delays in
   the middle of transmitting or receiving a message.  When the Host
   transmits a message on a link for which an entry is not in the link
   table, one will simply be inserted there.  There is no need for
   "dummy" Host messages to keep a link "open" since a link is
   effectively always open.  Only if the link table becomes full
   immediately after an entry is deleted (a situation we do not expect
   to occur) is there a possibility of resulting delay.

   Arbitrary delays introduced by Host programs are also not
   inconsistent with the link entry deletion procedure.  A link is
   blocked when the first access of the link table is made during
   transmission from the source IMP and is unblocked when the RFNM
   returns.  Only non-blocked transmit link entries are deleted after 30
   seconds of disuse.  The statement on page 23 referencing arbitrary
   delays was only intended to have hardware implications insofar as the
   Host/IMP interface is designed to transfer bits asynchronously
   between the Host and the IMP.

   A RFNM is returned from the destination IMP to the source IMP when a
   message reaches the head of the destination IMP's output queue to the
   Host (i.e. just before a message is sent to the Host).  If a
   destination IMP cannot then deliver that full message to the Host, at
   most one more message may possibly arrive at that IMP due to the
   premature release of the RFNM.  The new message will subsequently
   take its place at the end of the output queue to the Host thus
   guaranteeing the preservation of the proper message arrival sequence.

   The NOP message is a special control message which is available for
   use during initiation of communication between the Host and its IMP.
   The Host may, of course, decline to send NOP messages during this
   period, but the first received message after IMP startup or after the

   Host ready indicator has gone on, may be discarded by the IMP.  We do
   not require a Host to be prepared to repeat transmissions into the
   network.

   R.E. Kahn
   BOLT BERANEK AND NEWMAN INC.

         [ This RFC was put into machine readable form for entry ]
          [ into the online RFC archives by Marc Blanchett 3/00 ]