Uma Plataforma de Rede Definida por Software para Ambientes Convergentes de Computação Paralela

The growth in data volume has revolutionized business and science while demanding increasing capacity from computing resources. High performance computing platforms (HPC), traditionally employed in massively parallel numerical simulations, offer computational power that can be harnessed in big data analysis. However, the convergence of Big Data and HPC must be examined in several ways; In particular, network infrastructure needs to adjust to very different application demands. The software-defined network (SDN) model can favor this convergence, thanks to its global view of the network and its programmability. In this context, we present an SDN platform capable of convergently meeting Big Data and HPC application requirements. The platform applies routing mechanisms best suited to each traffic profile, thus reducing application execution time. We simulated the viability of our platform by reducing the execution time of real MPI applications in specific scenarios by up to 11% and Hadoop by up to 6%.