Parallelization of Array Method with Hybrid Programming: OpenMP and MPI

For parallelization of applications with high processing times and large amounts storage in High Performance Computing (HPC) systems, shared memory programming distributed have been used; a parallel application is represented by Parallel Task Graphs (PTGs) using Directed Acyclic (DAGs). the execution PTGs HPC scheduler executed two phases: scheduling allocation; considered an NP-complete combinatorial problem requires long times. Array Method (AM) to execute task schedule set clusters; this method was programmed sequentially, analyzed tested real synthetic workloads previous work. Analyzing proposed designs research work, extended hybrid OpenMP MPI server farm geographically at same time, novel for searching free resources clusters Lévy random walks proposed. Synthetic experimented evaluate performance new compare it sequential schedule. The metrics makespan, waiting quality assignments search were evaluated; results obtained described experiments section show better version algorithm compared version. By approach applied extraction characteristics PTGs, metaheuristic used, are improved. makespan decreased even when loads increase, tasks queue decreased, improved causing their subtasks be assigned or cluster neighbors and, finally, searches different clusters, not sequentially.