Synchronous and asynchronous implementations of relaxation algorithms for nonlinear network optimization (Q1180603)

Several kinds of physical systems have as model an optimization problem with network constraints and nonlinear costs. In this work, the authors present a parallel dual relaxation algorithm for network optimization. On a shared memory multiprocessor, synchronous and asynchronous implementations are used and alternative designs for parallel computing, that tradeoff synchronization delays with computations, are proposed. Empirically, that is to say by computational experiments, the performance is analyzed. Theoretical analysis and simulations indicated that the algorithm makes it a genuinely parallel optimization method.