CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > CFX

Minimum number of nodes to run CFX in parallel

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   April 6, 2005, 14:48
Default Minimum number of nodes to run CFX in parallel
  #1
Rui
Guest
 
Posts: n/a
Hi,

The CFX documentation (Solver Modelling, Using the Solver in Parallel, page 394) says that performance increase will only be noticeable if the number of nodes per partition is greater than 30,000 (for tet meshes) or greater than 75,000 (for hex meshes).

I'm doing transient free surface problems, for which I need quite small time-steps to achieve convergence. As a consequence of that, I have to use meshes with just a few thousand number of nodes. Even with these meshes, my simulations take some days (sometimes 1-2 weeks) to complete.

I allways run the simulations in only one PC with only one processor (Pentium 4). Do you really think that if I run a simulation, with a mesh made up of 2-3 thousand nodes, in parallel with 2, 3 or 4 PCs, I will not notice any performance increase as mentioned by the documentation?

Thank you for your help.

Regards,

Rui
  Reply With Quote

Old   April 6, 2005, 18:27
Default Re: Minimum number of nodes to run CFX in parallel
  #2
Glenn Horrocks
Guest
 
Posts: n/a
Hi,

The only way to be sure is to run it and try. I have got quite useful speedups on small meshes so there is a good chance you will go faster.

Glenn Horrocks
  Reply With Quote

Old   April 7, 2005, 20:58
Default Re: Minimum number of nodes to run CFX in parallel
  #3
Neale
Guest
 
Posts: n/a
You can try it but it will probably be minimal.
  Reply With Quote

Old   April 11, 2005, 20:46
Default Re: Minimum number of nodes to run CFX in parallel
  #4
Robin
Guest
 
Posts: n/a
Hi Rui,

Technically, you can run with as many computers as you have nodes, so a mesh comprising of a single hex element can be run on 8 cpu's! However, in practice the inter-process communication time begins to outweight the computation time as your node count drops.

That is not to say you will not see speedup. Generally, parallel efficiency will begin to drop off below 100k nodes and become significantly inefficient below 30k nodes. At 2 to 3 thousand nodes, you might still see some speedup, but probably nowhere near the number of cpu's (i.e. 4 cpu's will not give you a 4x speedup).

Best regards, Robin
  Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
correction number and parallel computing psosnows OpenFOAM Bugs 5 June 17, 2013 20:56
Unable to run OF in parallel on a multiple-node cluster quartzian OpenFOAM 3 November 24, 2009 14:37
Call for help! Cfx error in Parallel run. yq11ys CFX 4 July 27, 2009 10:09
Interpolation (Number of mapped nodes) Sam CFX 3 January 6, 2008 20:51
Cant run in parallel on two nodes using OpenMPI CHristofer Main CFD Forum 0 October 26, 2007 09:54


All times are GMT -4. The time now is 22:16.