|
[Sponsors] |
User routines cannot be used in the large memory partitioner? |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
March 13, 2017, 00:19 |
User routines cannot be used in the large memory partitioner?
|
#1 |
New Member
Tony
Join Date: Mar 2016
Posts: 24
Rep Power: 10 |
I would like to setup user routines in CFX. But it is written in CFX Help that user routines cannot be used in the large memory partitioner. So what is the so-called large memory partitioner? Is it the number of partitions while defining Run? What is the maximum partitions if use rountines are used?
Thanks in advance, Tony |
|
March 13, 2017, 05:44 |
|
#2 |
Super Moderator
Glenn Horrocks
Join Date: Mar 2009
Location: Sydney, Australia
Posts: 17,872
Rep Power: 144 |
When you do a multiprocessor simulation it has two steps:
1) Partition the mesh into the required number of partitions 2) Solve it by sending each partition to a core. Step 1 requires the entire mesh to be in memory for it to be partitioned. If your model has billions of nodes that is a challenge. So the large model partitioner is used to help here. See the documentation for further details. If you have enough memory to partition your model as it is then do not use the large model partitioner. It is only used when you run out of memory. |
|
March 13, 2017, 06:26 |
|
#3 |
New Member
Tony
Join Date: Mar 2016
Posts: 24
Rep Power: 10 |
Thanks, Horrocks.
The mesh nodes number is about 6 million and I got sufficient memory for this job. I go back to read the Help, but I am not sure if my understanding is correct. I have 28 cores on my computer, so I had better to select 28 partitions to make a parallel computation as fast as possible? If I just select one partition, then there would be only one processor joining the job? As shown in the help documents, if user routines cannot be used in the large memory partitioner, I have to define my run which incorporates user routines with small memory partitioner and the computation speed would be slow. Do I understand it correctly? Thanks in advance. |
|
March 13, 2017, 06:33 |
|
#4 |
Super Moderator
Glenn Horrocks
Join Date: Mar 2009
Location: Sydney, Australia
Posts: 17,872
Rep Power: 144 |
Do not assume that 28 partitions will run fastest. I would test 1, 2, 4, 8, 16, 28 and you will probably find the performance decreases once the partition count gets large. Designing compute clusters which run well with large number of partitions is a challenging area.
If you don't need the large partitioner (and for a 6M node mesh you shouldn't) then just use the default partitioner. You appear to have misunderstood - the large memory partitioner is only to be used when you run out of memory on the default partitioner. |
|
March 13, 2017, 06:36 |
|
#5 |
New Member
Tony
Join Date: Mar 2016
Posts: 24
Rep Power: 10 |
Thanks, Horrocks.
|
|
Tags |
cfx, partitioner, user routines |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Memory error? | Young | CFX | 3 | September 30, 2008 12:33 |
User fortran routines, DEBUG AND OUTPUT | Bloshchitsyn Vladimir | CFX | 5 | October 29, 2007 02:31 |
particle user routines | Felix | CFX | 1 | July 24, 2006 13:32 |
User routines | sac | CFX | 1 | December 8, 2005 11:42 |
CFX-5.7.1 MPICH distr. parall. + user routines | Claudia | CFX | 2 | March 18, 2005 12:39 |