Parallelization strategies for a reduced dimensionality calculation of quantum reactive scattering cross sections on a hybercube machine

The calculation of the reactive properties of complex chemical systems is a key step of the modelling of several modern technologies. These calculations are, in general, based upon the integration of the stationary Schrödinger equation for the motion of the atomic nuclei under the effect of the potential generated by the motion of the electrons. Related numerical procedures are highly demanding in terms of computer time and memory space. The goal of our work has been the restructuring of these procedures for implementation on a nCUBE 2 parallel computer. This paper, after a short presentation of the nCUBE 2 machine, describes two parallelization strategies adopted to obtain the best solution in terms of scalability and resources utilization with respect to the machine configuration. In the first parallelization strategy,the hypercube nodes have been clustered and every cluster is managed according to a processor farm model. In the second parallelization strategy, a single processor farm has been implemented on the host computer while all the hypercube nodes act as workers. Within each cluster one node acts as a farmer of the cluster. An analysis of the performance of the two solutions is presented.