Fig. 7-2 The computational grids for the coupled-model computation of a wing in the air

Fig. 7-3 Assignment of the three codes to the vector-parallel computer VPP300 and the scalar-parallel computer SR2201

Fig. 7-4 Improvement of the computational efficiency The flow of the simulation and the total computational times are shown in (a) and (b) for the cases when the codes are assigned to the most suitable computers among VPP300 and SR2201, and when the simulations are carried out by only one computer among them. The total computational times for the above three cases are compared in (c). The numbers of processors assigned in these cases are shown in parentheses. By assigning the codes to the most suitable processors the total computational time is considerably reduced.

Advancement of parallel computers enables us to carry out gigantic computations in various scientific and engineering fields, which have been previously unimaginable. One of important issues to be solved is how to efficiently carry out a simulation of a phenomenon represented by combination of several different models. If we calculate different parts of the phenomenon corresponding to the different models by different parallel computers most suited to the model, an even higher efficiency of the computation may be expected to be attained. When we carry out a calculation on a single serial computer, data are only generally processed within a single processor. In a parallel computer, on the other hand, computational data are exchanged among many processors and for this purpose the use of some data exchange aids such as the MPI library becomes necessary. Moreover, when we carry out a numerical simulation by making efficient use of different kinds of parallel computers connected through a network, a standard procedure is required for exchanging data among the different computers, and we have developed a new data-exchange library STAMPI for such a purpose. We confirmed that the STAMPI library can be used effectively for a real-world problem represented by a combination of different models. A concrete example of the coupled model problem is a simulation of a wing in the air, for which a fluid dynamics code, a grid generator, and a code for the structural mechanics are coupled (Fig. 7-2). For the fluid dynamics code and the grid generator, a vector-parallel computer, is most suitable. A scalar parallel computer can, on the other hand, work efficiently for the calculation of the structural mechanics. By using the STAMPI library we assigned the fluid dynamics code and the grid generator to the vector-parallel computer VPP300 and the code of the structural mechanics to the scalar parallel computer SR2201 (Fig. 7-3). In this way it has been confirmed that higher computational efficiency could be attained by cooperation among different types of parallel computers.

Reference T. Kimura et al., Local Area Metacomputing for Multidisciplinary Problems: A Case Study for Fluid/Structure Coupled Simulation, Proc. of the 1998 Int. Conf. on Supercomputing, Australia, 149 (1998).

Select a topic in left column