A semi-greedy heuristic for the mapping of large task graphs
Résumé
In the context of the dataflow process network mapping problem, we propose a parallel solutionmethod based on a semi-greedy heuristic to improve the solution quality of a greedy regret-based mapping algorithm (RBA). In this method, several executions of a randomized version of RBA are performed in parallel togenerate different solutions. We exploit a correlation between solution quality and execution time to propose a methodologyto stop the method after only a small amount of solutions are obtained.