|
[Sponsors] |
November 24, 2015, 07:08 |
MPI slower than serial
|
#1 |
New Member
Luca
Join Date: Jun 2015
Posts: 5
Rep Power: 11 |
Hi guys,
i'm trying to run my first parallel job on OpenFoam (but it's not my first time in parallel programming so i know the basic concepts). I'm working on a duct so i'm pretty sure that the best choice to reduce communications between processors is to divide the duct in parts along the flow direction. Therefore in the decomposeParDict i choose the simple algorithm with (4 1 1) coefficients (i'm working on a quad-core computer). The first problem is that the during the simulation one processor is faster than the others and waits for its neighbours only in the write times. But the serious problem is that the serial simulation is faster than the parallel one. I tried to run renumberMesh and got a little improvement but it's still slower than the serial. I'm working with more than 600k cells. Any guess about the reason why this happens? I just missed a flag!! [SOLVED] Last edited by LucaFen; November 24, 2015 at 10:45. Reason: I just missed a flag!! [SOLVED] |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
mpirun, best parameters | pablodecastillo | Hardware | 18 | November 10, 2016 13:36 |
parallel run is slower than serial run (pimpleFoam) !!! | mechy | OpenFOAM | 18 | August 17, 2016 18:19 |
Help: Serial code to parallel but even slower | Zonexo | Main CFD Forum | 4 | May 14, 2008 11:26 |
Error using LaunderGibsonRSTM on SGI ALTIX 4700 | jaswi | OpenFOAM | 2 | April 29, 2008 11:54 |
Is Testsuite on the way or not | lakeat | OpenFOAM Installation | 6 | April 28, 2008 12:12 |