Numerical recipes in C (2nd ed.): the art of scientific computing
Numerical recipes in C (2nd ed.): the art of scientific computing
A fully implicit Navier-Stokes algorithm using an unstructured grid and flux difference splitting
Applied Numerical Mathematics - Special issue: a festschrift to honor Professor Robert Vichnevetsky on his 65th birthday
| Hi-index | 0.00 |
The parallel performance of a large eddy simulation computational fluid dynamics code was evaluated on a Pentium III PC cluster with 64 processors running under Linux. Scaling capability of the code and comparative code performance under Fast Ethernet, Myrinet, and Gigabit Ethernet were investigated. The code shows very good scalability for a scaled speedup test with wall clock time remaining practically constant for any number of processors. The code does not saturate the Fast Ethernet bandwidth. Whereas Myrinet demonstrates a lower difference between CPU and wall clock time, indicating more efficient interprocessor communication, its overall wall clock time is equal to or higher than for the code compiled for Fast Ethernet. Gigabit Ethernet shows a very slight advantage over Fast Ethernet.