Set up for High performance cluster

c0sk · November 5, 2013, 17:06

Hi! I have 2 question regarding the flow past a sphere I am simulating in 3D, the sphere Diameter is 0.1m, nu=1.5*10-3 and is in a free stream U (4.5 0.1 0.1). With this set up (Re=300) will I be able to see Karman vortices using icoFoam?

Furthermore, to run this case I have access to a a ~6000 core high bandwidth, low latency Linux cluster of 2.66 GHz Intel "Westmere" cores based on 2 sockets and 12 cores per node. I have now idea what the optimal decomposePardict would be, I went through dozen of posts and manuals but still cant figure it out. Assuming I would use 6 nodes for a total of 12 sockets and 72 cores, what would I specify in the dict, which is the important value? 12 or 72?
Thanks
Regards
Andrea

jherb · January 30, 2014, 18:09

You probably found this out yourself: If you want to use 72 core you have to split your fluid domain into 72 partitions (and run 72 mpi processes)

joshm · January 31, 2014, 00:33

In regards to the first part of the second question, I think it would be highly dependent on the specific hardware. Because your calculations are iterating thousands of times to get your solution, small differences in the hardware's ability to perform a small task are greatly amplified. If you are going to be using the hardware for a long time, it's best to perform a benchmark test of your own.

November 5, 2013, 17:06	Set up for High performance cluster	#1
c0sk New Member Join Date: Aug 2013 Location: Italy/UK Posts: 24 Rep Power: 13	Hi! I have 2 question regarding the flow past a sphere I am simulating in 3D, the sphere Diameter is 0.1m, nu=1.5*10-3 and is in a free stream U (4.5 0.1 0.1). With this set up (Re=300) will I be able to see Karman vortices using icoFoam? Furthermore, to run this case I have access to a a ~6000 core high bandwidth, low latency Linux cluster of 2.66 GHz Intel "Westmere" cores based on 2 sockets and 12 cores per node. I have now idea what the optimal decomposePardict would be, I went through dozen of posts and manuals but still cant figure it out. Assuming I would use 6 nodes for a total of 12 sockets and 72 cores, what would I specify in the dict, which is the important value? 12 or 72? Thanks Regards Andrea

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Urgent problem! Appreciate all you help!! 3D Centrifugal Pump set up problems!	RR2	FLUENT	5	April 13, 2012 09:17
Parallel rerun in cluster	Andy_bm	OpenFOAM Running, Solving & CFD	4	November 27, 2011 08:16
How to show the transient case?	H.P.LIU	Phoenics	7	July 13, 2010 05:31
OpenFOAM installation on SUSE92 problems	derath	OpenFOAM Installation	1	April 3, 2006 07:44
link to cluster performance study	fourier	Main CFD Forum	0	March 8, 2002 02:00

January 30, 2014, 18:09		#2
jherb Senior Member Joachim Herb Join Date: Sep 2010 Posts: 650 Rep Power: 22	You probably found this out yourself: If you want to use 72 core you have to split your fluid domain into 72 partitions (and run 72 mpi processes)

January 31, 2014, 00:33		#3
joshm New Member Josh Join Date: Jun 2013 Posts: 19 Rep Power: 13	In regards to the first part of the second question, I think it would be highly dependent on the specific hardware. Because your calculations are iterating thousands of times to get your solution, small differences in the hardware's ability to perform a small task are greatly amplified. If you are going to be using the hardware for a long time, it's best to perform a benchmark test of your own.