Performance Analysis of a Skyline Solver on a Distributed Memory Parallel Supercomputer
Abstract
<div class="htmlview paragraph"><i>The performance of a parallel skyline solver is characterized analytically based on the average bandwidth, interprocessor communication speed, and the arithmetic processing speed. The formulas developed constitute a good predictor of the actual performance when the solver runs communication bound. This is the most interesting case because operation in this mode occurs for the largest processor configurations, and determines the ultimate performance of the solver. The analysis clearly shows the limiting effects of small-bandwidth coefficient matrices, and the relationship of processing speed and interprocessor communication bandwidth to the global performance. It is also shown that the largest potential gainsfor next generation machines will come from the availability of faster inter processor communication, rather than from enhanced arithmetic capability</i>.</div>
Related Papers
- → GPU acceleration of an unmodified parallel finite element Navier-Stokes solver(2009)60 cited
- → MPI-Based PFEM-2 Method Solver for Convection-Dominated CFD Problems(2022)6 cited
- → Large Scale 3D Multi-Phase-Field Simulation of Microstructure Evolution Using TSUBAME2.5 GPU-Supercomputer(2014)1 cited
- → THE PROBLEMS OF MODERN SUPERCOMPUTER APPLICATIONS IN HYDRODYNAMIC AND AEROACOUSTIC NUMERICAL SIMULATIONS(2010)
- → Creating and Using Solvers in the Openfoam Package for Modeling the Temperature Field(2023)