Network Working Group Jon Postel
Request for Comments: 346 Computer Science
NIC : 10425 30 May 72
Categories : Echo Plex, Satellite
References : RFC's 1, 5, 51
The consideration of using space satellite transmission links in the
ARPANET should be cause for some thought by the parties making use of
the network. The satellite transmission path will not necessarily affect
the transmission rate but it will affect the delay. The change in the
delay characteristics can be approximated by the change in path length.
Thus if the satellite is in synchronous orbit about 22,000 miles above
the earth, the path length is about 44,000 miles compared to (worst
case) 3,000 miles or about a 15 to 1 increase in path length and delay.
(The time for light to travel 3,000 miles is .016 seconds, to travel
44,000 miles is .236 seconds.)
In the current (surface) ARPANET delays are such that interactive
servers with character-at-a-time remote echo are only marginally useful.
While I believe that this delay (unmeasured) is largely due to the host
systems, adding a half second transmission delay will cause these
marginal systems to become unusable.
Thought should also be given to buffer allocations. If a receiving
system allows only one line of text to be buffered at a time and
refreshes the allocation as each line is output to a human user there
will be at least a half second delay between the arrival of each line at
the receiving system. This need not be a problem until the speed of the
output device is above about 150 characters/second. This "small buffer"
problem can be expected to occur even with lower speed devices since
host delays are estimated to be in the range 0.1 second to 1.0 second.
I suggest that it is appropriate to resume a discussion of measures to
circumvent the difficulties brought about by these large delay
characteristics. Some areas of discussion could be: buffer sizes in
servers and users, echo plex techniques, moving part of the input
processing to the user system. If it is decided to move the echo plex
functions to the user system, it would be wise to try for a "standard"
package, thus reducing a M times N problem to a M plus N problem.
Please dig out and read RFC's #1 Crocker, #5 Rulifson, #51 Elie to see
some previous thinking about this type of problem.
[ 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 ]