An Adaptive Random Search Alogrithm for Optimizing Network Protocol Parameters

The optimization of network protocol parameters based on network simulation can be considered a “black-box” optimization problem with unknown, multi-modal and noisy objective functions. In this paper, an adaptive random search algorithm is proposed to perform efficient and robust optimization for the concerned problems. Specifically, the algorithm is designed for use by the on-line simulation scheme which attempts to automate network management with protocol parameter tuning. The new algorithm takes advantage of the favorable statistical properties of pure random search and achieves high efficiency without imposing extra restriction, e.g., differentiability, on the objective function. It is also robust to noises in the evaluation of objective function since no traditional local search technique is involved. The proposed algorithm is tested on some classical benchmark functions and its performance compared with some other stochastic algorithms. Finally, a real case test is presented, in which the parameters of some RED queues are optimized with our algorithm.