questions on parallel efficiency

learner1111 · April 26, 2017, 04:33

Dear Support,

I've tested several cases, and I found that calculation is faster on my workstation than on our HPC platform. Same setup, same # of CPU, same MPI tool.

I don't have the exact hardware configuation at hand. But both hardware are new.

And, I see speedup with increasing # of CPUs on my work station. But on HPC, calculation first speeds up then slows down with increasing # CPUs.

can you give me some thoughts?

Thanks

Yaju · April 26, 2017, 14:25

Hi,

Does your HPC platform have the same number of cores per node as your local work station?

If not there can be slow down due to communication between multiple nodes.

You should also try your case with a large problem size (smaller base mesh size), and monitor your cells on each processor to make sure the load balancing is satisfactory.

MFGT · April 26, 2017, 15:27

Back in the days when I wrote my master thesis I ran simulations on a cluster, which had 8 cores per node. I guess this have been normal i7 processors, so they had hyperthreading on and simulation with 16 cores was just barely faster than local on 4 cores.

April 26, 2017, 04:33	questions on parallel efficiency	#1
learner1111 New Member Join Date: Feb 2017 Posts: 5 Rep Power: 9	Dear Support, I've tested several cases, and I found that calculation is faster on my workstation than on our HPC platform. Same setup, same # of CPU, same MPI tool. I don't have the exact hardware configuation at hand. But both hardware are new. And, I see speedup with increasing # of CPUs on my work station. But on HPC, calculation first speeds up then slows down with increasing # CPUs. can you give me some thoughts? Thanks Last edited by learner1111; April 26, 2017 at 06:42.

April 26, 2017, 14:25		#2
Yaju Member Yajuvendra Shekhawat Join Date: Mar 2017 Location: Convergent Science, Madison WI Posts: 51 Rep Power: 9	Hi, Does your HPC platform have the same number of cores per node as your local work station? If not there can be slow down due to communication between multiple nodes. You should also try your case with a large problem size (smaller base mesh size), and monitor your cells on each processor to make sure the load balancing is satisfactory. __________________ Yajuvendra Shekhawat Research Engineer - Applications Group CONVERGECFD

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[OpenFOAM] Questions about Paraview to show Parallel run of OpenFOAM	padian	ParaView	20	September 24, 2018 12:52
problem during mpi in server: expected Scalar, found on line 0 the word 'nan'	muth	OpenFOAM Running, Solving & CFD	3	August 27, 2018 04:18
Parallel Efficiency	rmojgani	FLUENT	1	December 5, 2012 04:31
Questions about Boundary Layer Thickness and Turbulence Models	famerfamer	STAR-CCM+	3	July 12, 2012 09:47
Parallel Computing Classes at San Diego Supercomputer Center Jan. 20-22	Amitava Majumdar	Main CFD Forum	0	January 5, 1999 12:00

April 26, 2017, 15:27		#3
MFGT Senior Member Tobias Join Date: May 2016 Location: Germany Posts: 284 Rep Power: 11	Back in the days when I wrote my master thesis I ran simulations on a cluster, which had 8 cores per node. I guess this have been normal i7 processors, so they had hyperthreading on and simulation with 16 cores was just barely faster than local on 4 cores.