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

from hardware to cells

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

Reply
 
LinkBack Thread Tools Display Modes
Old   April 13, 2014, 03:30
Default from hardware to cells
  #1
Member
 
Derek Mitchell
Join Date: Mar 2014
Location: UK, Reading
Posts: 95
Rep Power: 4
derekm is on a distinguished road
This is the reverse of the usual question, i have the configuration already determined, the question how many cells can it cope with, aiming for a 24hr run to simulate 1 to 3 hrs. The problem space is the heat flow in insect colonies using CHTmultiregionfoam or CHTMultiregionSimpleFoam.

The config is 4 servers with dual opteron 8381 (shanghai quad core 2.5GHZ) with 2 servers with quad 8381 opteron, total 64 cores and total 64Gb memory (800mhz). Gigabit interconnect.
__________________
A CHEERING BAND OF FRIENDLY ELVES CARRY THE CONQUERING ADVENTURER OFF INTO THE SUNSET
derekm is offline   Reply With Quote

Old   April 13, 2014, 04:36
Default
  #2
Super Moderator
 
Bruno Santos
Join Date: Mar 2009
Location: Lisbon, Portugal
Posts: 8,301
Blog Entries: 34
Rep Power: 84
wyldckat is just really nicewyldckat is just really nicewyldckat is just really nicewyldckat is just really nice
Greetings Derek,

Well, the proportions of "mesh size to RAM" may vary depending on the simulation software and the cell types in question. Because sometimes it's the vertex count that matters, others it's the face count and others is the cell count. Then it depends on the number of equations to be solved, if there are special mesh interfaces, how many subdomains there are (usually 1 subdomain per processor) and so on.

In your case, it's OpenFOAM and my guess is that the mesh is made mostly of hexahedral cells... although an ant colony isn't exactly easily meshed with a purely hexahedral mesh...
The rule of thumb I found a few years ago was that for the "cavity" tutorial, using blockMesh and icoFoam, the proportion seemed to be 1 GB of RAM for each 1 million cells. But as mentioned above, it's a purely hex mesh and a rather simple solver.

Therefore, in your case, my suggestion is that you should do a quick study of mesh proportion to occupied RAM, using the 1 GB/Mcell as reference for building 2 cases, e.g.: one with 16 million cells and another with 32 million cells. Then launch each case for a couple of iterations, to see how much total RAM it uses in each one.

Best regards,
Bruno
wyldckat is offline   Reply With Quote

Old   April 13, 2014, 18:03
Default
  #3
Member
 
Derek Mitchell
Join Date: Mar 2014
Location: UK, Reading
Posts: 95
Rep Power: 4
derekm is on a distinguished road
looks like 1.5e9 bytes per 1e6 cells for the chtmultipleregionfoam.

the next thing is to fathom a reasonable value per cpu
__________________
A CHEERING BAND OF FRIENDLY ELVES CARRY THE CONQUERING ADVENTURER OFF INTO THE SUNSET
derekm is offline   Reply With Quote

Old   April 15, 2014, 02:36
Default
  #4
Member
 
Derek Mitchell
Join Date: Mar 2014
Location: UK, Reading
Posts: 95
Rep Power: 4
derekm is on a distinguished road
Looking at the posts it seems 50k cells is a minimum number of cells per core and 2 core per numa is probably optimal. this would give 50k x 16 x 2
or 1.6E6 so that means the range would be 1.6E6 to 40E6 cells. with the processors being 2009 era i'm thinking closer to the 1.6E6 boundary.
__________________
A CHEERING BAND OF FRIENDLY ELVES CARRY THE CONQUERING ADVENTURER OFF INTO THE SUNSET
derekm is offline   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
No layers in a small gap bobburnquist OpenFOAM Native Meshers: snappyHexMesh and Others 2 November 25, 2012 09:54
Import netgen mesh to OpenFOAM hsieh Open Source Meshers: Gmsh, Netgen, CGNS, ... 32 September 13, 2011 05:50
snappyHexMesh won't work - zeros everywhere! sc298 OpenFOAM Native Meshers: snappyHexMesh and Others 2 March 27, 2011 21:11
snappyHexMesh aborting Tobi OpenFOAM Native Meshers: snappyHexMesh and Others 0 November 10, 2010 04:23
physical boundary error!! kris CD-adapco 2 August 3, 2005 00:32


All times are GMT -4. The time now is 06:07.