Parallel Jobs

Zaktatir · January 21, 2011, 11:43

Do not run small jobs in parallel

I am Calculating Multi phase Problems but using less than 100K Cells... if i am only calculation serial the simulation lasts a long time..

i discovered meanwhile sometimes big differences (especially if flow patter change occurs in my stratified flow problem) between serial run and parallel run.

Somebody have a hint or something...
thanks your Jessica

ghorrocks · January 22, 2011, 05:03

Of course there are time differences, you run parallel to run faster......

Can you explain what you are asking?

Zaktatir · January 24, 2011, 03:28

very smart of you. Off Course i know that the Simulation is faster. What i have meant is the differences in the results: i get another Flow regime for example.

It is not a question anymore of the time duration but whether is the solver too sensible for parallel Computing in the case of very less number of cells (less than 100K cells Hexa). In the documentation it stays: never run small jobs on parallel..

But if my physics are almost complex (Multiphase flow) running it serial is very time expensive...

ghorrocks · January 24, 2011, 05:49

Getting a difference between a serial and parallel simulation is rare. There are a few cases listed in the documentation of when it can occur, such as when a free surface aligns with a partition boundary for instance. But often you can re-partition to avoid this problem and it works fine.

Can you describe what you are modelling?

Have you tried different partitioning options to try to stop partition boundaries aligning with flow features?

Zaktatir · January 24, 2011, 06:58

I am using HP MPI Distributed Parallel partitioning algorithm, and this is recommended in the Documentation..
Yeah i am Modeling stratified flow (Air-Water) in channel (inclined part elbow and horizontal)...
Entrainment may and would occur for high gas velocities.

I am noticing this differences especially in high gas velocities regime.

ghorrocks · January 24, 2011, 16:33

MPI is not the partitioning algorithm, it is the run mode. You need to change the partitioning technique.

In solver manager, select "Show Advanced Options", then "Partitioner". The default partitioner type is metis. Change it to something else and try again.

Zaktatir · January 25, 2011, 03:12

Sorry my mistake..

Is there any recommandation for using partitioning algorithm for multiphase flows (wavy stratified including entrainment and separation) or shall i try every kind of included algorithm in CFX

ghorrocks · January 25, 2011, 05:22

There is nothing magical about the different partitioning techniques - they just give you different partition shapes. The key thing is if a partition boundary aligns with a partition boundary then for a small range of simulations this can cause problems. So just about ANY different partitioning technique will move the partitions boundaries to different spots and probably fix the problem.

But the best solution is to read the documentation and understand what the different techniques actually mean. Then you will be able to choose the best one for your application.

January 21, 2011, 11:43	Parallel Jobs	#1
Zaktatir Senior Member Join Date: Jan 2010 Location: Germany Posts: 268 Rep Power: 17	Do not run small jobs in parallel I am Calculating Multi phase Problems but using less than 100K Cells... if i am only calculation serial the simulation lasts a long time.. i discovered meanwhile sometimes big differences (especially if flow patter change occurs in my stratified flow problem) between serial run and parallel run. Somebody have a hint or something... thanks your Jessica

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
CFX parallel jobs on LSF	Dominik	CFX	4	June 22, 2022 13:18
Script to Run Parallel Jobs in Rocks Cluster	asaha	OpenFOAM Running, Solving & CFD	12	July 4, 2012 22:51
multiple parallel jobs on one machine	joeybernard	CFX	0	December 16, 2010 10:10
running multiple Fluent parallel jobs	Michael Bo Hansen	FLUENT	8	June 7, 2006 08:52
How to distribute no. of CPUs among parallel jobs?	SGI Altix	CFX	3	February 14, 2006 01:38

January 22, 2011, 05:03		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Of course there are time differences, you run parallel to run faster...... Can you explain what you are asking?

January 24, 2011, 03:28		#3
Zaktatir Senior Member Join Date: Jan 2010 Location: Germany Posts: 268 Rep Power: 17	very smart of you. Off Course i know that the Simulation is faster. What i have meant is the differences in the results: i get another Flow regime for example. It is not a question anymore of the time duration but whether is the solver too sensible for parallel Computing in the case of very less number of cells (less than 100K cells Hexa). In the documentation it stays: never run small jobs on parallel.. But if my physics are almost complex (Multiphase flow) running it serial is very time expensive...

January 24, 2011, 05:49		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Getting a difference between a serial and parallel simulation is rare. There are a few cases listed in the documentation of when it can occur, such as when a free surface aligns with a partition boundary for instance. But often you can re-partition to avoid this problem and it works fine. Can you describe what you are modelling? Have you tried different partitioning options to try to stop partition boundaries aligning with flow features?

January 24, 2011, 06:58		#5
Zaktatir Senior Member Join Date: Jan 2010 Location: Germany Posts: 268 Rep Power: 17	I am using HP MPI Distributed Parallel partitioning algorithm, and this is recommended in the Documentation.. Yeah i am Modeling stratified flow (Air-Water) in channel (inclined part elbow and horizontal)... Entrainment may and would occur for high gas velocities. I am noticing this differences especially in high gas velocities regime.

January 24, 2011, 16:33		#6
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	MPI is not the partitioning algorithm, it is the run mode. You need to change the partitioning technique. In solver manager, select "Show Advanced Options", then "Partitioner". The default partitioner type is metis. Change it to something else and try again.

January 25, 2011, 03:12		#7
Zaktatir Senior Member Join Date: Jan 2010 Location: Germany Posts: 268 Rep Power: 17	Sorry my mistake.. Is there any recommandation for using partitioning algorithm for multiphase flows (wavy stratified including entrainment and separation) or shall i try every kind of included algorithm in CFX

January 25, 2011, 05:22		#8
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	There is nothing magical about the different partitioning techniques - they just give you different partition shapes. The key thing is if a partition boundary aligns with a partition boundary then for a small range of simulations this can cause problems. So just about ANY different partitioning technique will move the partitions boundaries to different spots and probably fix the problem. But the best solution is to read the documentation and understand what the different techniques actually mean. Then you will be able to choose the best one for your application.