A descent algorithm for the multihour sizing of traffic networks

Multihour engineering is a technique for designing trunk networks when the hours of peak traffic loads between various pairs of offices do not coincide. A new descent-type computational algorithm for the multihour engineering problem is derived. This algorithm obtains the unique solution to the minimization of the multihour cost function, which is strictly convex but only piecewise differentiable. The noninteger minimum-cost solution is subsequently rounded to the nearest allowable integer solution to give a realizable network. The new algorithm is applied to three numerical examples from the California network. The results are compared with the nonoptimal, nonunique solutions obtained with an earlier algorithm, and with the traditional single-hour solutions.