Optimal placement of extrapolators for random flow prediction (Q801799)

The article investigates the quality of a linear predictor of the sum of random flows as a function of the placement of extrapolators in the network elements. The problem of optimal flow partitioning into groups is formulated in terms of minimum mean square prediction error. The problem is presented as a nonlinear integer program. A theorem concerning the structure of the optimal programs is proved, and this theorem is used to design an efficient algorithm for optimal extrapolator placement.