IMPLEMENTATION OF A DECENTRALIZED PARETO-OPTIMAL ALGORITHM

iterative flow control algorithms to achieve a Pareto optimal point have been proposed in the telecommunications literature. In this paper, we present several relaxation and estimation techniques to improve the performance of the algorithms. We also explore several possible implementations that extend support to the proposed analytical models. Simulations of these possible implementations are carried out, and the performance of the flow control algorithms studied. The proposed implementations for the Pareto algorithms achieve optimization of the product of powers of individual user classes. The effect of additional factors like the length of the observation cycle, the structure of the observation cycle, and the relaxation factors are also studied. The simulation results suggest that better convergence is achieved when the observation windows are maintained at the node level and are packet count based. The window size has to be sufficiently long. The relaxation techniques are useful with the right choice of relaxation factors. The greedy algorithms using estimation techniques to achieve complete decentralizability are very appealing.