THROWDOWN MACHINE FOR TRAFFIC STUDIES 297 



based on a knowledge of the statistical behavior of the various factors 

 entering into the composition of real traffic. Random numbers are drawn 

 for each factor. These numbers are assigned values according to fre- 

 quency distributions obtained analytically or from field observations. 

 The regulating data are combined to produce a description of a sample 

 of traffic which would be encountered under the assumed conditions and 

 then processed by methods which simulate the performance of the actual 

 system. 



As a simple example of the throwdown procedures, suppose that it 

 is desired to determine how often on the average an ''all trunks busy" 

 condition will occur in a particular group of trunks handling inter-office 

 calls. A certain period of time is first selected and the number of calls 

 expected within this period is determined. Two random numbers are 

 then drawn for each call. One random number specifies the time, within 

 the period, of origination of the call. The other random number, weighted 

 according to an exponention distribution which will be discussed later, 

 gives the holding time of the call. 



With the data of call origination times and holding times prepared, 

 the throwdown run can be started. The calls are listed in the order of their 

 originating times. The first call is assigned to the first idle trunk. A record 

 that this trunk is busy is made and the time at which it will become idle 

 determined by adding the assigned holding time to the time of origination. 

 This is also recorded. The call which follows in time of origination is then 

 assigned to the next idle trunk and the process continued for succeeding 

 calls. Before each call is established the release times of all busy trunks 

 are scanned to determine whether any busy trunk should be made idle. 

 In setting up each call idle trunks are chosen from the group in the same 

 order of preference that would be used in the system being simulated. 



Thus, the performance of an actual system is reproduced with con- 

 siderable accuracy and detailed records of this performance can be made. 

 From a study of these records the desired information can be determined. 

 The probability of encountering an "all trunks busy" condition can be 

 found, the average number of trunks busy can be determined, or a 

 frequency distribution chart showing the percentage of the time the 

 number of busy trunks is above any given number can be constructed. 

 If proper records are kept such information as the average number of 

 trunks searched over in locating an idle trunk can be determined. If the 

 trunks were reached through a graded multiple or if they were in sub- 

 groups with a common overflow group, simple extensions of the above 

 procedures would be foUoAved. 



This particular problem can be solved by analytical methods and is 



