THROWDOW^ MACHINE FOR TRAFFIC STUDIES 309 



inasmuch as subscriber behavior is very slightly influenced by the type 

 of system ser\dng their telephones. 



Subscribers, although they are indi\dduals, exhibit many ''group char- 

 acteristics" dictated not by the requirements of telephone conmiunica- 

 tion but by their mode of life. This fact allows statistical treatment of 

 many observed action distributions Avithout introduction of significant 

 error. However, these group actions also present problems of congestion 

 in telephone plant which require detailed throwdown study for solution. 



As an example of group characteristic, subscribers do not originate a 

 steady barrage of calls over the twenty-four hours of the day. During 

 mid-morning and mid-afternoon hours traffic is built to a peak value, 

 whereas during certain of the remaining hours it is reduced to a mini- 

 mum. In some residential areas peak traffic may also occur during the 

 early evening. Throwdown evaluations of simulated smtching systems, 

 however are primarily concerned with the busy hour, the hour in which 

 the greatest number of calls are originated, regardless of its actual time 

 of day occurrence. 



Useful datum obtained from busy hour field observations is the calling 

 rate per subscriber (calls per hour) which can be used to set up traffic 

 load conditions on the simulated smtching system. The calling rate 

 characteristics can be measured as average calls per hour placed by 

 subscribers in a number of group classifications. An example used in a 

 particular throwdoA\Ti study of simulated system response to a given 

 traffic load is given in Table I. The values given in this table represent 

 average day to day calling rates. Weather conditions, pre-holiday peri- 

 ods or special events have been found to raise substantially the average 

 calling rate in affected classifications. Values adjusted for these condi- 

 tions are useful in projecting percentage of overload that can be offered 

 to systems engineered for average daily loads. 



Subscribers, however, in originating calls, act independently Av-ithin 

 their classified group in maintaining the average calling rate. Originating 

 times of calls, therefore, occur at random A\dthin the hour. ThrowdoA\Ti 

 input data representing subscriber originating time behavior are produced 

 by assigning to each call, of the total Anthin the studied hour, a six digit 

 number from a list of random niunbers. If the hour is divided into one 

 million parts the assigned random number determines the miUionth 

 part of the hour in which each call aaiII originate. 



Observations made in the field have shoAATi that subscribers, upon 

 receiving dial tone, do not always follow through to dial a full code. 

 Among possible causes are failure to hang up after completion of a call, 

 answering the wTong telephone where two or more are adjacent, diahng 



