Estudio del overhead en la migración de algoritmos paralelos de cluster y multicluster a GRID

An experimental work is presented, in which a multicluster application has been migrated. In this application, communications among processes/processors was handled directly from MPI to the same architecture with Grid support (Globus Toolkit 4.04) in order to study the overhead generated by the incorporated middleware. The N-queens solution with static distribution has been chosen and two clusters interconnected with Grid support, including the GridFTP, RFT, GRAM, WS-GRAM, RLS, and MDS services, were configured. Later, the response time for the N variable was studied, analyzing the speedup and efficiency. The most interesting result is the overhead independence generated by the Grid middleware when scaling the problem.