A Comparison of Two Master-Worker Scheduling Methods

The mathematical frameworks of two schedulers, SCOW and UMR, are used to tune up parameters that optimize the amount of communications and parallel computations in a single-program multiple-data parallel code, under certain constraints. Comparisons are made with simulated data that is fed into the mathematical models, and with performance data taken from the scheduling of a parallel method for finding a string motif in a family of DNA sequences. The latter compares also the make-spans predicted by the underlying mathematical models with the actual results. Some considerations on the validity of the theoretical frameworks and the potential for a hybrid SCOW-UMR scheduler are presented, as well.