TOWARD RELIABLE OPERATION

OF MINICOMPUTER-BASED TERMINALS ON A TIP

       The present protocol for communication between a TIP and
attached terminals requires character-oriented transmission and
provides for no error control. In the design of this protocol, it was
apparently assumed that the majority of terminals attached to a TIP
would be interactive, be normally used in a character-by-character
mode both for transmission to and from the terminal, and normally
support a human user who would in effect be in the communication loop.
The human user would thus be in a position to detect any significant
telecommunication-induced errors both by direct observation of the
character stream and, more importantly, by examining the computer
output in the context of his ongoing interaction.

       The effectiveness of this means for error detection and
initiation of corrective measures when necessary is not adequate in
the following cases:

a. For terminal-TIP communication at a medium or
higher data rate (say 1200 bps or higher) it is quite possible that the human will skim computer output and not be an effective character-by-character error detector. In particular, when both user input and computer output contain numerical data it is possible that significant undetected errors could occur.

b. For terminals located at a distance from the TIP and connected either by a private line or the switched network more errors may be introduced than with a terminal local to the TIP (see Note 1). When a large
number of user terminals are connected to TIP's through telecommunications facilities, whether within a single organization or, even more likely, when users and user groups not needing the full TIP capability are connected
to a remote TIP, this problem may arise.

d. An interesting potential application of the network is to provide support for minicomputers used for process
control and other laboratory measurement functions. In providing software support for such minis as well as acquiring data from them usually there is no human user
in the communication loop.

e. A number of sites already offer a remote job entry service. Although the present sites assume that the unit record devices such as card readers and line printers are files within a multiprogrammed system at another site, it appears natural that remote batch terminals be attached
to the network through TIP's. Here again, there would be no human in the loop between the terminal and the TIP.

       In addition to some degree of error control on these types of
terminal loops, it may be desirable to provide for block-oriented data
transmission, at least for terminals of types (d) and (e) and possibly
(c) above. It is possible that error control utilizing block
transmission can be superimposed on the present TIP-terminal
communication protocol. Data blocks, including error control and block
delimiting information, can be multiples of a single character in
length. The communication channel would still not be as fully utilized
as for conventional synchronous block communication, since start and
stop bits for each character would need to be transmitted. This loss
is not substantial and does occur now for 2000 bps TIP-terminal
communication.

         There are at least two ways to implement such a protocol on
top of the existing TIP-terminal communication protocol. In both
cases, the remote terminal would have to handle both originate and
receive error and block control procedures:

a. Through an addition to TIP software, the controlled communication loop could terminate in the TIP, thus providing error control only where it is most needed, between the TIP and the terminal. This, however, would involve additional TIP software and a block buffering capability which may put an excessive load on the TIP.

       If the remote end of the block transmission error control loop
is in the serving host, then this software could possibly be used for
host-to-host, end-to-end error control in addition to
host-host-terminal end-to-end error control. For host-to-host
communication, however, there would be a slight loss in efficiency due
to the imbedded character-oriented format, unless an option were
provided in which start/stop bits were not required.

-------------------------------
Note 1: The most recent published data concerning data transmission
error performance of the switched telecommunications network is
provided in the 1969-70 Connection Survey conducted by Bell
Laboratories. The results are published in The Bell System Technical
Journal, Vol. 4, No. 50, April 1971.  In this survey, 12 receiving and
92 transmitting sites in the U.S. and Canada were used with standard
Bell System Dataphone datasets used at both ends.  At both 1200 and
2000 bps, approximately 82% of the calls had error rates of 1 error in
10^5 bits or better, assuming an equal number of short, medium, and
long hauls.

       The results of this survey for low-speed, start/stop data
transmission at rates up to 300 bps indicate a character error rate of
1 error in 10^4 characters or better on 77.6% of all calls made within
the survey. It is interesting to note that only 48.3% of the low-speed
data tests completed were error-free. These tests were nominally 40
minutes in length.

       For voice grade private line data channels, the Bell System
technical reference, "Transmission Specifications for Voice Grade
Private Line Data Channels," dated March 1969 reports "When a Bell
System dataset is combined with the recommended channel, the expected
long term average error rate of the system is 1 error in 10^5 bits or
better during normal transmission conditions. "

       [ 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   ]