On Distributed Communications Series
IV. Priority, Precedence, and Overload
II. Traffic Overload
A communications network can carry only a finite volume of traffic; if this volume is exceeded, a degradation in performance will result. If a local "step-by-step" civilian telephone office is overloaded and unable to accept a call, the busy signal is immediately heard. If the terminating central office is busy, the busy signal is returned shortly after the number is dialed. Usually, this delay is annoying but not critical. One can wait and try again.
Commercial communications networks are designed around an underlying assumption that each telephone is probably used less than two per cent of the time. A small amount of switching equipment can be safely shared by a large number of intermittent users.
It has been found, historically, that telephone subscribers are well satisfied if they are able to pick up the telephone and obtain service within ten seconds 99 per cent of the time. This level of service is called the probability time, P-T (0.01, 10); that is, the probability that service will be unavailable after ten seconds will occur only about one per cent of the time for any subscriber.
An additional doctrine of commercial common-carrier switched networks is that every subscriber shall receive the same grade of service.
Ninety-nine per cent service may sound like excellent communications service. But, to the military user such a network leaves much to be desired. The user has no choice as to exactly which one per cent of the time service is to be denied. The one per cent failure time must be expected to occur when there is an abnormally heavy demand upon the network. In civilian networks this usually happens during unexpected snowstorms, widespread fires, hurricanes, floods, etc. This can be tolerated since the goal of the commercial telephone utility is to provide service at the lowest cost most of the time; it is not basically intended for general emergencies.
Military crises, almost by definition, place abnormal loads on systems. When using a communications system designed under such civilian loading assumptions for military purposes, one can expect most service denials to occur precisely during those times when most needed. Though this may sound like a recitation of the obvious, the implications of such underlying ground rules have, on many occasions, been unappreciated by some planners of systems for the military.
This common, implicit, system-design assumption is particularly treacherous, as there is little opportunity to discover its existence under normal test operation. Communications networks are rarely exercised in real-time to simulate extensive communications network damage and overload.
Overloading is one of the causes of breakdowns in store-and-forward systems during military crises. During periods of high tension, not only does command-control traffic increase, but even logistics traffic is generated in greater volumes. In a crisis almost everyone feels obliged to communicate: the heavy increase of low-priority logistics traffic has been called the "underwear ordering" effect. The crisis will evoke a flood of backlogged requisitions into the system, all demanding immediate processing.
A significant increase can usually be accepted by the local communications tributary station for processing; the bottleneck occurs farther downstream. Present-day "hard copy" written-text military communications networks are slow-speed store-and-forward systems. Long-time intermediate storage is used at the switching nodes to improve high-cost long-line-circuit usage. When the traffic volume arriving at the intermediate switching center from the many feed points is greater than the output circuit can handle, messages must be backlogged.
There are several different military precedence systems in use, which, theoretically, insure that the more urgent traffic is processed first. For example, in the BIX, Binary Information Exchange, system the precedence categories of "Flash," "Emergency," "Operational Immediate," "Priority," "Routine," and "Deferred" are used. These categories are divided into two subsets, "High Precedence" and "Low Precedence," each handled by a different set of rules within the store-and-forward switching center.
The earlier UNICOM system originally considered only three grades of precedence, "Right-of-Way," "Priority," and "Routine"; more recently this program has incorporated the use of four precedence grades. The Defense Communication Agency has standardized on the set shown in Appendix A.
Each user of the conventional military store-and-forward system is responsible for suitably marking his own messages fed into the network. The military communicator's goal is to process all highest-priority traffic so that it is delivered to the addressee within X minutes of transmission; all of the next grade traffic within Y minutes; etc. Since these times cannot be met under heavy traffic conditions, a downgrading procedure, called "Minimize," is called into play (see Appendix B). "Minimize" attempts to reduce the volume of high-precedence traffic. Although helpful, the procedure has not been wholly effective in reducing network overload. The writer has heard it said that the following chain of events occurs: Messages inserted into the network are delayed enroute by a time factor unknown to the originating party. Not having any confirmation of receipt of his urgent message, the originator panics and sends another message--at an increased precedence level. Still not receiving an answer, the originator again panics and dumps still more high-precedence traffic into the network.
This mechanism may be viewed as being akin to a fastacting servo controller connected in a loop with a long signal feedback lag-time; oscillation can be expected. A communication network that does not let the user know how long it will be before his message will be delivered to the end addressee may be theoretically oscillatory.
A similar overload mechanism is sometimes observed in the civilian telephone system during unexpected civil crises. A caller dials a number and hears a busy signal. Upon receipt of the busy signal, he dials again. Again, hearing the busy signal, he becomes more impatient and once more ties up the shared central office equipment. In extreme cases, the telephone company may evoke a doctrine called line-load control. This consists of intermittently cutting off blocks of subscriber lines feeding the overloaded offices. Line-load control not only provides better service to the remaining users, but safeguards against a complete central office failure by overload of the timeshared common-control markers used in some dial systems. Such panic-induced failures, tying up an entire central office for several hours, have been noted. (The telephone utility in its efforts to maximize continuity of service in the face of overload, calls a series of separate techniques into play as the load builds up near the critical point; these are described in Appendix C.)
 R. I. Wilkerson's comments to M. Juncosa and R. Kalaba, in "Optimal Utilization of Trunking Facilities," Communications and Electronics, No. 40, January 1959, p. 1002.