The Multi-point Approximation Method (MAM) replaces an optimization problem
by a sequence of approximate ones. The corresponding approximate response
functions are simple and often explicit in terms of the design variables. E
ach step of the optimization process involves several implicit function eva
luations before identification of the approximation functions by weighted l
east-squares fitting takes place. The present paper addresses alternative p
lanning of implicit function evaluations in order to enhance both robustnes
s and efficiency in parallel computer environments, especially heterogeneou
s networks.