DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 365 



parities measured along the delay axis in the higher ranges of the variable, 

 are, of course, considerably less. A comparison of the theoretical and 

 observed proportions of calls delayed, and the average delays on all calls 

 is shown in the following table: 



These differences between theory and observation are well within the 

 variations which would be expected with the lengths of throwdown 

 runs made. 



Further reassurance that the traffic submitted in the two throwdown 

 tests originated in a manner reasonably similar to that assumed in the 

 theory was obtained by making ''switch counts" at regular intervals 

 during the throwdowns from which frequency distributions, /(x), of 

 the number x of calls simultaneously present were constructed. These 

 are sho\vn in Figs. 5 and 6 for the two throwdown cases. The solid 



0.10 



Z0.09 

 to 



< 0.07 



<0.06 

 o 



g0.02 



II 



X 0.01 



16 



CALLS CALLS 

 BEING WAITING 

 SERVED 



20 24 28 32 36 40 44 48 52 



NUMBER OF CALLS BEING SERVED OR WAITING 



Fig. 5 — Distribution f{x) of simultaneous calls. Theory versus throwdown, 

 c = 2 paths, « = 0.90, 3000 throwdown calls. 



