Estimation of Point-to-Point Telephone Traffic

Estimates of point-to-point telephone traffic are required for the current and the long-range planning of the Bell System's Public Switched Network. Because of the potentially immense volume of data which must be processed, these estimates are typically based upon small samples of total traffic and, therefore, can have large statistical errors. In this paper, we develop a model for quantifying the accuracy of point-to-point traffic measurements as a function of sample size and traffic parameters. Together with a worth-of-data model, not described here, our results can be used to establish a cost-optimal sampling rate for point-to-point traffic measurement systems. However, our results have been used to establish 20 percent as an upper bound on a cost-optimal sampling rate for a usage measurement system and 10 percent for an attempt-only measurement system. We show, however, that the attempt-based estimate is, for sampling rates greater than about 2 percent, less accurate than the usage-based estimate. We also show how the accuracy of point-to-point load estimates can be improved by employing a ratio-estimate which combines point-to-point and trunk-group measurements; however, in practical applications, we find that the improvement is not significant.