Network Working Group                                       David Walden
Request for Comments: 534                                        BBN-NET
NIC: 17453                                                  17 July 1973
References:  512, 516, 533


                        Lost Message Detection

  As an aside to RFC 533, note that if sending Hosts do uniquely
  identify messages on a given link using the extra four bits and
  receiving Hosts do look at these bits, a lost message detection
  system such as those suggested in RFCs 512 and 516 drops right out of
  using of the unique message-id.  These extra four bits can be treated
  as Hathaway's SCB of RFC 512 providing a 16 element sequence number
  on a per connection basis.  A 16 element sequence is sufficient as
  the IMPs never allow more than four outstanding messages at one time
  between a given pair of Hosts.  As Hathaway also suggests, the 0
  element in the sequence can be used to indicate to the receiving Host
  that sequence numbers are not being used.

  To summarize, there appear to be three modes of using the message-id
  number under Host/Host protocol:

  1. The sender can always set the extra four bits to 0 and only
     transmit one message over a given link at a time -- this is slow
     but it allows orderly retransmission of messages without any help
     from the receiver.

  2. The receiver can give no help to the sender.  In this case it
     doesn't matter whether the sender uses the extra four bits to
     uniquely identify the messages or not -- the sender has no method
     of orderly retransmission, although the sender can accurately
     identify which message was lost if the sender has uniquely
     identified the messages.

  3. The sender can have multiple messages outstanding (i.e., RFNMs not
     received) on a given link and the receiver can help the sender.
     In this case, if the sender uses the extra four bits to uniquely
     identify the messages in a way which can be synchronized with the
     receiver (e.g., sequential id numbers), the receiver can reliably
     detect lost messages.

  Although it probably will seem insufficient to some, if the sender
  and receiver use synchronized unique message-id numbers, very
  reliable retransmission schemes are readily available.  For instance,
  the sender can retransmit the appropriate messages in response to
  incomplete transmissions and the receiver can use the unique
  message-ids to sort the retransmitted messages into the proper order



Walden                                                          [Page 1]

RFC 534                  Lost Message Detection             17 July 1973


  with the other received messages.  Alternatively, the receiver can
  discard all messages received out of order and the sender can back up
  and retransmit a message for which an incomplete transmission was
  received and all subsequent messages.


        [ This RFC was put into machine readable form for entry ]
      [ into the online RFC archives by Alex McKenzie with 10/99 ]











































Walden                                                          [Page 2]