THROWDOWX MACHINE FOR TRAFFIC STUDIES 295 



system performance. The level and character of submitted traffic can 

 be varied, and a wide range of system sizes with varying quantities of 

 control circuits can be tested. 



Before the throwdown machine was built a ''paper" throwdown trial 

 of a small No. 5 crossbar system installation had been conducted by 

 Mr. R. I. Wilkinson. This was run by a team of girls using card files, 

 ledgers and written records, and using dice to make certain random 

 decisions. The machine is basically a mechanization of these early 

 methods to make possible the testing of larger installations in a reason- 

 able time. The methods of generating data for the machine were devel- 

 oped by Mr. Wilkinson and many of the decisions relating to telephone 

 traffic and statistical problems which were encountered in designing the 

 machine were solved in consultation Avith him. 



THE TELEPHONE TRAFFIC PROBLEM 



A large automatic telephone switching system of the common control 

 type is not a simple mechanism nor is evaluating its performance and 

 traffic canying capacity a simple problem. The economy of these systems 

 depends upon the efficient use of relatively small groups of circuits on a 

 time-sharing basis to serve a large number of subscribers. Each group of 

 circuits is specialized to perform certain of the functions necessary in 

 establishing a connection, and circuits from several groups must co- 

 operate to handle every call. A sequence of actions with appropriate 

 alternatives at several stages where busy conditions may be encountered 

 is completely prescribed for every call. However, this sequence is subject 

 to interference due to simultaneous requests to use the same control 

 circuits. Competition is resolved by preference arrangements which 

 cause some rec[uests to be delayed while others are being served. Delays 

 will increase the holding time of circuits with the possibihty of causing 

 traffic congestion at other points in the system. 



With a number of subscribers originating calls at random, it becomes 

 difficult to predict what the reactions of the system mil be at various 

 traffic levels. Although some parts of the problem can be solved by 

 analytical methods employing probability theory, it is doubtful that 

 mathematical means, beyond rough approximations, are available for 

 evaluating the performance of an entire system. 



Where systems have been built and placed in operation, the per- 

 formance can be judged from observations of the working system. There 

 are obvious weaknesses in this procedure. Only by collecting large 

 quantities of information can the performance at various load levels 



