RELIABILITY OF TRAFFIC MEASl^REMENTS 3G9 



Equation (12) may also be deri\'0(l with the ))1'(»('(m1ui-(' used for eciuation 



(9) using al = al -{- al . Foi' ^•aluo,s of /' large enougli to make ctnhf - j 



= 1 e(iuation (12) is approached by equation (9). For smaller values 

 of /• (but with r still great(M- than 1), 1% for constant holding times is 

 less than 1',. for exponciilial holding times. When r = 1, there is no 

 carried load error. For \'alues of r less than 1, the coefficient of varia- 

 tion of error in estimating source load average will vary from 



depending on the exact \'alue of r. It is interesting to note that Vs for r 

 = 0.5 is the same as for r — 1.125. 



b. Loss 



The effect of loss in the group depends upon the disposition of the lost 

 calls. In general, accuracy in measuring carried load increases with in- 

 creased loss because under these circumstances fewer load changes occur 

 l)(Hween scans. This is evident in the extreme case of a group which is 

 100 per cent loaded; a single switch count gives a correct reading for any 

 length period. Obviously load readings at 100 per cent occupancy are 

 not very useful in estimating offered loads since the amount of lost load 

 cannot even be guessed at. However, in the cases of lost calls held 

 (Poisson) or cleared (Erlang B), the offered load may be estimated from 

 the carried load (less and less accurately as occupancy increases) and in 

 the case of lost calls delayed the offered and carried loads are likely to be 

 the same even at high occupancies. With high loss, therefore, estimates 

 of source load are subject to errors not considered in deriving ecjuation 

 (99); however, switch count error in estimating carried load will be 

 materially less than predicted by equation (1). 



c. Random Call Originaiion 



On trunk groups which are alternate routes, calls may no longer be 

 considered as oi-iginating at random. The resultant grouping of call orig- 

 inations will tend to dcci-ease the accuracy of switch count measui'ements 

 in estimating carried load; however, there is a corresponding decrease in 

 accui'acy in estimating the source load from the carried load so that ac- 

 curacy in estimating carried load may be less worthwhile. 



