[kinakamichibun]

When I tried to run the tutorial of refiningMovingCylinderInChannelIco of foam -extend 4.1 in parallel, it firstly aborts for two undefined keywords, they are numberOfSubdomains and method. But after these two keywords are added, it still aborts after several iterations with the following message:


Fatal error in MPI_Recv: Message truncated, error stack:
MPI_Recv(200).....................: MPI_Recv(buf=0x114f910, count=49, MPI_BYTE, src=0, tag=1, MPI_COMM_WORLD, status=0x7ffd2df53490) failed
MPIDI_CH3U_Receive_data_found(131): Message from rank 0 and tag 1 truncated; 51208 bytes received but buffer size is 49
This tutorial can run well in serial.Now I don't know what to do with it. Can anybody give some advice? Thank you!

[AliE]

Originally Posted by kinakamichibun

When I tried to run the tutorial of refiningMovingCylinderInChannelIco of foam -extend 4.1 in parallel, it firstly aborts for two undefined keywords, they are numberOfSubdomains and method. But after these two keywords are added, it still aborts after several iterations with the following message:


Fatal error in MPI_Recv: Message truncated, error stack:
MPI_Recv(200).....................: MPI_Recv(buf=0x114f910, count=49, MPI_BYTE, src=0, tag=1, MPI_COMM_WORLD, status=0x7ffd2df53490) failed
MPIDI_CH3U_Receive_data_found(131): Message from rank 0 and tag 1 truncated; 51208 bytes received but buffer size is 49
This tutorial can run well in serial.Now I don't know what to do with it. Can anybody give some advice? Thank you!

Hi, I am not familiar with openFOAM extended, however this error looks nasty, since it comes from mpi_recv and thus it is located inside the information exchange between processors.

If you are not familiar with mpi, the error reported means that the proc0 is sending a message of a certain size but for some reason the prescribed receiver is not able get it correctly. This might be caused by an incorrect tag or an incorrect receiver id into the mpi_recv call or an incorrect memory allocation of the buffer array.

Since the code comes from the extended version and the serial version is running well, there is a good probability that you have a bug in the sources... Sorry, but it is diffucult to be more helpful than this.

[LuckyTran]

This isn't a problem of immersed boundary layer method but a how to run openfoam in parallel question. Posting the mpi error msg is not at all helpful, what you need is the actual openfoam log from the host-node (the one that looks like the same stuff you'd see in a serial message).

Are you able to run any cases of OF in parallel? I think the answer is no.

In decompseParDict you have to specify numberOfSubdomains and method. Then you need to run decomposePar. After you do this you will find dirs like processor0, processor1, and so on corresponding to the same number that you specified in numberofSubdomains. Are you here yet?

Then you need to launch OF in parallel mode, how to do this depends on the environment. For example, in my case, I have to invoke mpirun.

[AliE]

Originally Posted by LuckyTran

This isn't a problem of immersed boundary layer method but a how to run openfoam in parallel question. Posting the mpi error msg is not at all helpful, what you need is the actual openfoam log from the host-node (the one that looks like the same stuff you'd see in a serial message).

Are you able to run any cases of OF in parallel? I think the answer is no.

In decompseParDict you have to specify numberOfSubdomains and method. Then you need to run decomposePar. After you do this you will find dirs like processor0, processor1, and so on corresponding to the same number that you specified in numberofSubdomains. Are you here yet?

Then you need to launch OF in parallel mode, how to do this depends on the environment. For example, in my case, I have to invoke mpirun.

Yes, this is another option. I assumed in my answer that you have already done the procedure suggested by luckyTran So hopefully you have missed some steps.

[kinakamichibun]

Originally Posted by LuckyTran

This isn't a problem of immersed boundary layer method but a how to run openfoam in parallel question. Posting the mpi error msg is not at all helpful, what you need is the actual openfoam log from the host-node (the one that looks like the same stuff you'd see in a serial message).

Are you able to run any cases of OF in parallel? I think the answer is no.

In decompseParDict you have to specify numberOfSubdomains and method. Then you need to run decomposePar. After you do this you will find dirs like processor0, processor1, and so on corresponding to the same number that you specified in numberofSubdomains. Are you here yet?

Then you need to launch OF in parallel mode, how to do this depends on the environment. For example, in my case, I have to invoke mpirun.

Definitely I know how to run foam in parallel, actually I seldom do serial simulations. I have run several cases in parallel with this installation, so problems should not come from my installation. The error message of undefined keywords of numberofsubdomains and method is located at dynamicMeshDict file, which is somewhat interesting

[vesp]

Originally Posted by kinakamichibun

Definitely I know how to run foam in parallel, actually I seldom do serial simulations. I have run several cases in parallel with this installation, so problems should not come from my installation. The error message of undefined keywords of numberofsubdomains and method is located at dynamicMeshDict file, which is somewhat interesting

well, if you rule out ansinput errorson your part, it is likely a bug in the code. Just work it out likeyou would any other bug...
If I had a dollar for every time I ran into an MPI issue, I would be rich. Just dig in and get dirty, or talk to support if you are not a programmer.

[LuckyTran]

What even is openfoam extend 4.1? Can you send me the github link?

[akidess]

[removed...]