[alexksei]

Hi all

I am running simulations on quite large meshes (150-500 mln cells).

The questions are:

1. Is there any way (formulas, etc...) to estimate how much memory CFX will estimate for allocation for interpolator, partitioner, and solver?

2. I am running simulation on a cluster with several nodes. The problem is that CFX will perform partitioning step using just 1 node (and all the nodes for solving). The node has 64 GB, but CFX is estimating memory for allocation being 125 GB, which is 2 times more than the system memory. I tried different partitioning methods (METIS, Directional RCB, Optimized RCB, RCB) but with no success (maximum memory reducton was 30%). Is there any way to ask CFX to use several nodes for the partitioning step? Or any other suggestions of how to overcome this problem?

3. Another problem with interpolator step. CFX is estimating required memory being very small. I increase memory allocation factor (with a multiplier that would give almost 100% of available memory). But in the end I am getting the message that insufficient memory allocated. Though the requsted space is smaller than allocated space (see below). CFX is behaving very tricky and I just don't understand what to do.

+--------------------------------------------------------------------+
| *** INSUFFICIENT MEMORY ALLOCATED *** |
| |
| ACTION REQUIRED : Increase the real stack memory size. |
| |
| Details : |
| Requested space : 249402429 words |
| Current allocated space : 625000000 words |
| Current used space : 498805384 words |
| Current free space : 126194616 words |
| Number of free areas : 1 |
+--------------------------------------------------------------------+

[ghorrocks]

I understand CFX has been tested up to about 2 billion elements, so it should be able to handle your model size. The basics:
* I trust you are running this in CFX stand-alone, not in workbench. Running a challenging model like this in WB is just asking for trouble.
* Have you activated the "large model" option in CFX-Pre/Execution Control/Run Definition/Run Settings? You will need that.
* You will probably need to use recursive bisection rather than METIS for partitioning, but you are already doing that.
* I am not sure 64GB is big enough to do the partitioning. There is no work around here, if a single node cannot partition it then you are stuck.

[Gert-Jan]

Instead of cluster, use a large workstation with multiple processors and enough Gb (256 Gb?). Then for partitioning, the single processor can allocate all memory present in the workstation.