Solving Traveling Salesman Problem Using Parallel Genetic Algorithm and Simulated Annealing

The traveling salesman problem (TSP) is to find a tour of a given number of cities (visiting each city exactly once and returning to the starting city) where the length of this tour is minimized. In this project, we implemented solutions to the traveling salesman problem using parallel genetic algorithm and simulated annealing, and compared the parallel performance and results of these two implementations. In performance analysis, we used the results from running serial implementation of the two algorithms as baselines for comparison. Furthermore, we studied the effect of changing the rate of convergence on the parallel performance of these two algorithms. The two algorithms are implemented in C and parallelized using the MPI library. The study indicates that both the genetic algorithm and simulated annealing implementations achieve good parallel speedup. We found that increasing the rate of convergence improves the optimization result (prior to reaching a plateau) but also significantly increases the running time.