A Descent Algorithm for the Multihour Sizing of Traffic Networks

This paper describes a numerical algorithm which obtains the unique, optimal noninteger solution to the multihour traffic network engineering problem. This solution is subsequently rounded to the nearest allowable integer solution to yield a unique, near-optimal realizable network. As described in Ref. 1, multihour engineering is a procedure whereby a least-cost traffic network is engineered for more than one set of point-to-point loads, subject to the constraint that blocking on any last-choice trunk group not exceed a specified value. For networks which exhibit noncoincident traffic patterns,* the multihour engineering method has been shown to achieve significant capital-cost savings over the conventional single-hour engineering procedures.1 The results reported in Ref. 1 were based on an algorithm which optimizes the high-usage trunk group sizes* one at a time, in a fixed but t Traffic loads between different pairs of offices are said to be noncoincident if their highest average values occur in different hours, or at the same hour but in different seasons. 1 High-usage groups are direct groups which carry the majority of the load between those pairs of offices which have a large enough community of interest to warrant direct trunking. 1405