On Distributed Communications Series
IV. Priority, Precedence, and Overload
V. Media Control
To this point, we have described how available data rate, a volatile commodity whose measure is in bits per second, may be allocated among the many users of a communication network. We shall next consider a method to encourage each individual user to make most efficient use of his allocated communications resource.
The common-user digital network of the future encounters a wide variety of data-generating devices. These devices can load the network at rates differing as much as 100,000 to 1. Some of these devices are briefly described in Appendix D, and include telephone, teletype, telegraph, and facsimile. Each of these media places a different bit-rate load upon the communications system.
In the civilian world, the decision of choice of medium tends to be primarily economic. We ordinarily use the cheapest medium commensurate with the value and perishability of the information. In the non-dollar economy of the military world we need a mechanism analogous to the concept of money to encourage each individual user to select the medium which makes most efficient use of the total communications resource.
The reader is cautioned that the following development is not the result of rigorous analysis, inasmuch as the analytical tools are simply not available. But, in their absence, we have chosen the following approach; the numbers used are not important since their purpose is illustrative only.
Let us start with a tentative assumption that each alternative communications medium conveys roughly about the same amount of "information theory" type information per unit time under real-time, human-to-human, or human-to machine information interchange. This is not too difficult to visualize: For example, humans speak and type at the same rate, about 60 words per minute. But, in the information theory sense, such language is highly redundant. Tests have been run to determine the non-redundant information input rate of humans, and an extensive body of literature has developed examining the input data rate for man's input senses. The results of these tests indicate that the maximum possible human input data rate is probably less than about 50 bits/sec, based upon tachistoscopic picture element recognition (visual input), language text (also visual input), voice (auditory input), and vibratory Morse code (tactile input).
As "information" can be sent in a variety of ways, we wish to encourage each user always to use the most "efficient" medium. During peacetime, it might be most "efficient" to send a letter by facsimile if transmission bandwidth is cheap and keyboard operator's time expensive. However, during overload the available supply of the communications commodity decreases, and, assuming a free economy, the "price" should change.
Under these conditions, one would like to encourage the use of more-efficient, narrower-band transmission devices, such as, say, teletype. We would like to have a "cost" table of these different data sources in order to provide a basis for determining the inflation-free price of service when the communications resource is to be rapidly rationed (see Table 1).
Data rate alone, however, does not provide a complete measure of network loading; some devices have a short duty cycle, such as one computer sending the contents of its core to a remote computer. While such devices place a heavy peak demand for service, they are highly intermittent. On the other hand, a pulse-coded telephone call places a lower peak demand load, but ties up network capacity for a longer period and results in heavier average loading. Therefore, we should include an expected message-duration or holding-time factor in the network-load weighting table.
Two separate factors are at work here. Many separate low-data-rate devices time-shared or concentrated into a single high-data-rate link permit better averaging, as compared to a few correspondingly-higher-data-rate users. But, as many of the high-data-rate users "get in" and "get out" fast, they have a short holding time. This helps the averaging process. To be precise in this computation, a better understanding of the number of users, their use statistics, and the network characteristics appears mandatory, and shall be deferred until such information is available. It is sufficient for purposes of explaining the concept to use the following tentative rationale in preparing the sample loading chart of Table II.
Start with the data rate in bits per second of the transducers which we have briefly considered. Next, give "credit" to those devices expected to be in use for only a comparatively small portion of the time. Column 1 lists the peak data rate in bits/sec for each of the input devices. Column 2 lists the percentage of network users using any single type of data input device. (If, for example, ten per cent of the links served 40-bit/sec keyboards, the value of ten per cent would be shown in Col. 2.)
Column 3 shows the expected maximum duty cycle of each input device averaged over a long period. (A telephone could conceivably be used 100 per cent of the time, while it is doubtful that a computer core dump would occur more often than 0.1 per cent of the time.)
Column 4 is the product of data rate and users, and provides an indication of the loading demand by each type of input source. Column 5 lists the rank order of entries of Column 4 to indicate the type of devices that make heaviest average demands upon the data resource. Thus, for example, in our network we will expect many digital telephone users but few computer core dumps.
This determination of allocation of loading demand can be further refined by including the following two factors.
First, we could use the investment cost of each input device as a factor indicating its "importance." Thus, for example, another multiplicative factor could be included to allow for the higher investment cost of computers, relative to telephones.
Secondly, we could also use the reciprocal of the total number of each data input device as the metric. Using this rationale in a distributed command-control system, the true vulnerability is probably some function of the reciprocal of the number of times a specialized input device is replicated in the network. For example, if three special purpose computers are tied together with a communications network, destruction of a single computer installation (or its communications) is correspondingly more serious than if the complex contains fifty computers sharing the work. In the first case, there would be loss of one-third of the complex's capacity; in the second, only 1/50 capacity. Thus, we might wish to tend to favor the more "critical" elements as compared to the less unique.
The metric being developed as a measure of the communications resource is a vector having many components. To this point, we have described a few of the components. We shall now consider yet another component--the conventional military precedence indicator and how it might be blended into the vector. The present-day precedence indicator system concept is based primarily upon the speed of delivery of a message. Historically, it has probably grown out of the old commercial telegraph tariffs; i.e., telegram, deferred, day letter, and night letter.
Column 1 of Table III lists present Defense Communications System Precedence Categories, together with target processing times. Column 3 lists the approximate ratio of these time categories. An underlying consideration in the following development is the mixed requirement that, while we wish to give priority treatment to the higher-precedence traffic of equal network loading, we must also satisfy the goal that we preserve a minimum transmission capability for the lower-precedence traffic. Thus, instead of a blanket rule that all traffic of a given precedence grade will be transmitted before handling the next lower precedence grade, we choose to use the time ratios of these precedence categories to act as a preference weighting factor.
Table IV combines the network loading factors for different digital services, shown in Table I, together with the numerical values of preference weighting factor derived from Table II.
Table V is included to illustrate certain combinations of "lower-precedence" traffic which can automatically force preferential treatment over those forms of "high-precedence" traffic making extremely inefficient use of network data rate. The entries of Table V are listed in order of preference. This illustrates how we can encourage each network user to make an efficient choice of the form of data transmission, regardless of his chosen precedence grade, without depriving any low-precedence user from transmitting a small volume of traffic which adds negligible loading to the network.