Section: New Results

Simgrid for MPI (SMPI)

Several new results on the usage of Simgrid to assess MPI performance have been published in 2017. The general framework introducing the methodology for a proper use of SimGrid to simulate MPI applications was presented in [5]. One more specific line of work concerns the prediction of the performance and the energy consumption of MPI applications using SimGrid [39], [19]. Other applications have also been simulated using this approach. General capacity planning of supercomputers is analyzed in [35] using simulation. More specifically, we have shown that the HPL benchmark (high Performance Linpack), used to establish the top 500 ranking of the most powerful supercomputers in the world, can be emulated faithfully on a commodity server, at the scale of a supercomputer [36]. We have also shown that SimGrid is reliable and fast enough to evaluate and tune the performance of dynamic load balancing in seismic simulations [22].