Algorithms for perishable inventory control have high computational requirements when solving large problems. Therefore, the use of High Performance Computing (HPC) reduces the execution time and improves the quality of the solutions. Here, parallel implementations of an optimization algorithm for perishable inventory control problems have been developed. Heterogeneous platforms, where every node is composed of several multicore processors and Graphic Processing Units (GPUs), have been considered. Developed parallel versions are: (1) a MPI-PTHREADS version, which takes advantage of the parallelism at thread and process levels; and (2) a multi-GPU version, in which process level parallelism and GPU core level parallelism are exploited. Experimental results show the benefits of using HPC to solve this kind of problems. Furthermore, the distribution of the workload among the available processing elements is a challenging problem. This distribution of tasks can be modeled as a Bin-Packing problem. This implies that the selection of the set of tasks assigned to every processing element requires the design of a heuristic capable of efficiently balancing the workload statically with no significant overhead. The selection of several heuristics is available in the implementations.
Autors: Alejandro Gutiérrez-Alcoba1, Gloria Ortega2, Eligius M.T. Hendrix1 and Inmaculada García1
1 Dpt of Computer Architecture, Escuela de ingenerías, c/ Dr Ramos, Universidad of Málaga, 29071 Spain.
2 Dpt Computer Architecture. Almeria University (ceiA3), Ctra Sacramento s/n Almeria 04120 Spain.
Contact: agutierreza@uma.es; gloriaortega@ual.es