Network Working Group                              Bernard Cosell
RFC # 271                                          BBN
NIC 7819                                           3 January 1972
Categories:  B.1
Updates:  None
Obsoletes:  None

                     IMP System Change Notification
                     ------------------------------
	

    We are planning to install a new version of the IMP System,
version 2514.  The new version is scheduled for field installation
Thursday, January 13, 1972 between noon and 1 PM EST, and will require
the assistance of IMP-site personnel at each site.


    There were two principal problems with version 2513, both related
to the delay inserted between the time when a Host comes up and the
time when its IMP will accept the second packet from the Host.  The
first was that the delay was lengthened to slightly over 40 seconds,
which caused hardware difficulties for some Hosts.  The second was
that there was an ambiguity that could make the delay run as long as a
minute and a quarter.  On the first point, the delay has been backed
off from 40 seconds to 30 seconds, as it was for IMP systems prior to
2513.  On the second point, the ambiguity has been entirely removed.
(Note, however, that BBN Report No. 1822, on page 23, specifies that
the delay may range from 30 to 90 seconds, and that future versions
may require a longer delay.)


    In summary, a Host may come alive in one of two ways, corres-
ponding to the two ways in which the Host can go down.  If the Host
went down voluntarily (by sending a "Host going down" to the IMP), the
Host indicates his intention to come alive by sending the IMP
something.  If the Host went down involuntarily (by dropping his ready
line), the Host indicates his intention to come alive by bringing his
ready line back up.  In either case

the IMP will refuse to accept more than one packet from the Host for
30 seconds after the Host has indicated his intention to come alive.
Notice, however, that the Host must be prepared to accept all messages
from the Network from the instant that he indicates his intention to
come alive.* This particular point seems to have given many Hosts
difficulty in running through their standard initialization
procedures.  Don't forget this simple and universal rule, that when
you are telling your IMP that you are alive, you must be prepared to
always take every- thing from the Network whether or not the Network
is taking any- thing from you.

    Version 2514 will also incorporate a few other changes, mainly
related to the operation of the NCC.  Since the Timeout is, for a
change, being made shorter, and the other modifications are minor,
there should be no appreciable transient with the coming of the new
version.

_______________
*In fact, if the Host does not accept messages from his IMP
pimmediately then the Host may see the IMP's Ready line go down
for 1/4 second sometime during the 30 second waiting period.
This is due to the following set of circumstances:

    *  The IMP periodically places NOPs on the queue for a
       dead Host as part of the process of checking the IMP/Host
       interface.

    *  If a message remains on a Host's queue for 30 seconds
       without being taken, the IMP drops its Ready line for
       1/4 second in order to clear the interface (see RFC #270).

    *  The timeout periods for the Host queue and the delay
       when the Host comes alive are _not_ synchronized.

If the Host is prepared to see the IMP's Ready line dropped
during the 30-second delay while coming alive, then no harm
will be done if messages from the IMP are not accepted immediately.


BC/jm

       [ This RFC was put into machine readable form for entry ]
       [ into the online RFC archives by BBN Corp. under the   ]
       [ direction of Alex McKenzie.                   12/96   ]