Methods for improving and evaluating the performance of unstructured CFD-algorithms

The problem of improving the performance of unstructured CFD-algorithms on multicore processors is discussed. The details of the application of algorithms for reducing the adjacency matrix bandwidth and the implementation of the calculation scheme with asynchronous time integration are considered. The possibility of using a set-associative cache simulator for a comparative evaluation of data ordering methods is shown. A procedure for testing the speed of a finite-volume algorithm for modeling the Navier-Stokes equations is described using the example of calculating the problem of supersonic flow around a sphere on three mixed meshes containing from 270 thousand to 17 million cells.