Parallel Processing

If DAO run times became excessive, parallel processing could be used to gather results of individual rounds in less overall elapsed time, allowing the best of the best to rise to the top more quickly. Initially configuration parameters for constraint values, dorm definitions, blocked beds, annealing options and data files would be distributed. On a per round basis, each slave process reports its total score whenever available, and the master checks if it is a new record score, and signals the slave to save the configuration if so, otherwise the master signals another round or the end. Since slave processes do not have to synchronize with each other, the varying times of the schemes are less of a problem.

download thesis