Visualizing the Performance of Scientific Applications Executing with Space-Time Domain Decomposition
Today’s high performance computers offer the possibility of computing complex large scale computational fluid dynamics (CFD) simulations in a reasonable time. In practice, exploting the thousands of computing cores to achieve the desired solution time is very challenging. These challenges include finding enough work to keep all these cores busy, and assigning work to different parts of the computing platform that can include different types of multicore processors and manycore graphics processors. A promising approach involves simultaneously computing different time intervals in the simulation, using what is called the parareal algorithm. Careful monitoring of the performance of the computation is required in order to fine tune the implementation of this algorithm so that it produces the desired benefit. In a recent internship a visualization tool was developed, which allows online monitoring of the progress of a CFD simulation. This project will investigate the further development and use of this tool to allow effective use of the parareal algorithm on current high performance computers. This will enable Envenio to achieve significant performance gains on large-scale real world fluid dynamics problems.