[lostking18]

Hi, there

Recently while I ran the simulation in CFX, the simulation always crashes with the following warning message：

###############################################
+--------------------------------------------------------------------+
| ERROR #001100279 has occurred in subroutine ErrAction.
| Message:
| Failed to allocate dynamic memory for non-blocking CNL communication.
+--------------------------------------------------------------------+
| ERROR #001100279 has occurred in subroutine ErrAction.
| Message:
| iif_create_empty: error allocating new index: malloc of 2560256 bytes failed
+--------------------------------------------------------------------+
| ERROR #001100279 has occurred in subroutine ErrAction.
| Message:
| iocnt: open the primary file failed
###############################################

I’m running the simulation with work station(Intel I9-18 core, 3.0GHz frequency, 128G Ram). The intel MPI local parallel with 20 partitions is used. This crash comes out almost randomly. Sometimes it works, sometime the simulation will crash after hundreds of iterations. Could you kindly tell me how can I improve my workstation or change the set-up of the solves to eliminate the sort of crash?

Thank you in advance

[ghorrocks]

I have no idea what is happening there, I have never seen it before. It is a CFX error message, so CFX is not allocating enough memory for something, but I do not know what CNL communication is. It is definitely something unusual, so there is something unusual about your simulation to trigger it.

All I can suggest is to do some basic debugging. Does it happen if you do the same simulation but with a smaller mesh? Have you tried all the memory options? eg Large problem solver and partitioner, memory allocation factors, and all the other options (both CEL and command line).

Does the problem occur when you run the same simulation but with less cores? (I note you have an 18 core machine and am running 20 partitions - this does not sound wise).

[lostking18]

Hi, Ghorrocks, thank you very much for your always helpful answers.

I already enact the "Large Problem solver". I'm running a turbine stage steady simulation with 15 cooling holes. The total number of mesh nodes is around 10million. There are plenty of interfaces between the cooling injection and the main flow domain.

Sorry that I'm not familiar with partition allocation. Do you suggest I should use 18 partitions instead of 20?

Originally Posted by ghorrocks

I have no idea what is happening there, I have never seen it before. It is a CFX error message, so CFX is not allocating enough memory for something, but I do not know what CNL communication is. It is definitely something unusual, so there is something unusual about your simulation to trigger it.

All I can suggest is to do some basic debugging. Does it happen if you do the same simulation but with a smaller mesh? Have you tried all the memory options? eg Large problem solver and partitioner, memory allocation factors, and all the other options (both CEL and command line).

Does the problem occur when you run the same simulation but with less cores? (I note you have an 18 core machine and am running 20 partitions - this does not sound wise).

[ghorrocks]

Yes, try 18 partitions.

How many domains are in your simulation?

[lostking18]

Originally Posted by ghorrocks

Yes, try 18 partitions.

How many domains are in your simulation?

17domains: 1 for stator, 1 for rotor, the rest 15 for the cooling blocks

[ghorrocks]

OK, that is not a huge amount of domains. So what have you tried so far?

Also, if you could post the output file that might help. No guarantees, this is weird one.

[Opaque]

Would you mind sharing which version you are using?

Very unusual error indeed. Could you also share when it did happen in the overall solution algorithm? Say, what is it when solving one of the equations set? Mass+Momentum, or Energy, or Turbulence?

Could you also post the memory summary in the output file? it should indicate how much of the machine memory is being used by the simulation alone, which will hint at how much memory is left for dynamic allocation later on.

By the way, you said you were running 20 partitions on 18 cores, correct? What was the goal of doing such partitioning?