Abstract
We propose a new efficient analytic-imitational method of optimizing non-Markov queueing networks. We estimate its convergence speed and precision with experiments and give practicai recommendations for its use.