Network Working Group John M. McQuillan
Request for Comments #410 Bolt Beranek and Newman
NIC #12402 10 November 1972
Categories: B-1
Removal of the 30-Second Delay When Hosts Come Up
-------------------------------------------------
The IMP currently delays accepting input from a Host for 30
seconds after the Host has come up. This delay serves to allow
the fact that the Host is up to propagate through the network.
The fundamental problem is that a Host must not be permitted
to communicate with a second Host until the second Host
(actually its IMP) has been made aware that the first Host is
up. Otherwise, one Host may come up and send a "hello"
message to another Host, whose reply is discarded by the IMP
because it is for a dead destination.
All this reasoning is based on a dead destination de-
tection mechanism at the source IMP. The 30-second delay is
based on the worst-case propagation delay for routing information
in the network, so that every potential source IMP can update
its host up/down table. There are several drawbacks to this
scheme:
1. Hosts should not have to wait the worst-case time
of 30 seconds to send to Hosts at their IMP or
nearby in the network.
2. The operation of half-duplex interfaces is made
even more complicated because of the startup delay.
3. The timeout period of 30 seconds is really a
function of network topology and we would like to
be able to change it when necessary as the network
eXPands.
We propose to eliminate the 30-second delay altogether.
The IMP subnetwork will detect messages for a dead Host at the
destination IMP instead of at the source IMP. There is no delay
[Page 1]
involved for an IMP to sense when its own Hosts come up, so
that it can always make the correct decision about whether to
give a message to one of its Hosts or to return a destination
dead message to the source Host. Under this new scheme, when-
ever the IMP's ready line is up it is ready to accept input
from its Hosts without delay. Several comments on this change
should be noted:
1. No change to Host software should be necessitated
by this change. The Host may attempt to send a
message to the IMP as soon as it brings its
ready line up, or it may delay for a long time. In
either case, the IMP will take the message. As
before, as soon as the Host has brought up its
ready line, it must accept messages from the IMP.
2. The Hosts may wish to remove any delays _they_ have
programmed into their startup routines, since
sUCh delays are no longer necessary.
3. Destination dead messages will be returned as
before with two differences. There will be more
delay between the receipt of the message at the
IMP and the return of the dead destination message
because it must travel through the network. For
the same reason, if many messages are sent to
dead Hosts, the dead destination messages may come
back out of order.
The Host personnel responsible for the IMP software at
each site should check that this proposed change will not ad-
versely affect them. If no adverse comments are received,
this change will go into effect on Tuesday morning, December
12 at the regular IMP software release time.
[ This RFCwas put into machine readable form for entry ]
[ into the online RFCarchives by BBN Corp. under the ]
[ direction of Alex McKenzie. 1/97 ]