[sidharath]

Hello all,

I was running a transient simulation using CFX on HPC and saved the results at every timestep.
Unfortunately, space on server ran out before and run couldn't finish.
To overcome this, a friend suggested me to not to save the output at all ("none") and instead monitor the variables/monitor points for the stably converged solution and then save that timestep manually.

This approach works just fine but the problem is I am running the simulation on HPC and can only access the .out file which is being updated as the run goes on and .res file will only be generated once the simulation is completed.
Can anyone help me with the way to monitor the run in real-time using only .out file?
Is there a better way to approach this problem?

Thank you and help would be much appreciated.

[ghorrocks]

Most simulations are monitored during progress with the output file. It is the normal way to do it. The CFX documentation describes the contents of the output file.

You should be looking at least at the convergence, time step sizes, any error/warning messages and the conservation at the end of the run.

[monkey1]

Hmm funny way of "saving" disk space.
I would rather suggest:
1. Why do you write a trn file EVERY time step?
2. For the trn-file settings choose not to write out ALL variables (as it is standard setting) but just selected variables you really need...this reduces the disk space the most.

When I look at my cases: allthough my .out files are huge (smth. around 100 mb) it is nothing compared to the corresponding .res (around 8 GB) and the Results folders...but there the trn. files are "only" 1GB each because I selected only some variables to be saved instead of "All".

[sidharath]

Originally Posted by ghorrocks

Most simulations are monitored during progress with the output file. It is the normal way to do it. The CFX documentation describes the contents of the output file.

You should be looking at least at the convergence, time step sizes, any error/warning messages and the conservation at the end of the run.

Thank you for the reply, Glenn. I am trying to monitor the user points along with convergence and imbalance. Unfortunately, when I am opening the .out file with CFX solver manager, I can only see the text of .out file but not the user points or any plot in the monitor of solver manager.
For elucidation, I've attached the picture of solver manager I am getting by opening the .out file as well as the one I get while I run the simulation on my local machine (with user point data).
Can you please suggest any way to get the monitor the user points with .out file in my case. I do get mon file while running the simulation on HPC and I tried to change the extension of mon file to .res file but no luck.
Any pointers would be really useful.

Thank you.

[singer1812]

Just monitor the run directory on your HPC.

Open up solver manager, select Monitor Run In Progress, navigate to your run directory (it should be a your_run_name_001.dir), then hit enter.

Solver manager should operate the same as on your PC. You should also be able to edit the case while it is running, same as on your PC.

[sidharath]

Originally Posted by monkey1

Hmm funny way of "saving" disk space.
I would rather suggest:
1. Why do you write a trn file EVERY time step?
2. For the trn-file settings choose not to write out ALL variables (as it is standard setting) but just selected variables you really need...this reduces the disk space the most.

When I look at my cases: allthough my .out files are huge (smth. around 100 mb) it is nothing compared to the corresponding .res (around 8 GB) and the Results folders...but there the trn. files are "only" 1GB each because I selected only some variables to be saved instead of "All".

Hi monkey1 (sorry),
Thank you for the tips on saving the space but my predicament is I am not sure about what all parameters I would need to look at for my problem (first time with this problem for me) and hence can't make a certain choice to ignore/not save some variables.
I am setting up an improvised case to save the output every 1 revolution of the wheel(72 timesteps; 5 deg/timestep). To be honest, I am still studying the sensitivity of timestep and I am also not sure about the number of revolution it would need to get stable/periodic results.
Can you suggest any way that would consume less space for the case of time step size independence study?
Thank you.

[sidharath]

Originally Posted by singer1812

Just monitor the run directory on your HPC.

Open up solver manager, select Monitor Run In Progress, navigate to your run directory (it should be a your_run_name_001.dir), then hit enter.

Solver manager should operate the same as on your PC. You should also be able to edit the case while it is running, same as on your PC.

Hi singer1812,

As per my understanding, I don't think I can navigate to run directory as it is over an external network. It also needs usual authentication to access the terminal and hence, when I try to select the run directory in the solver manager I can only see/navigate to local machine folders.
Is there a way around it, to link up the HPC server to local computer so that CFX solver manager can monitor the run?

Thank you.

[singer1812]

You would need to talk to your HPC people on this. I have no idea how they have it setup.

How are you able to see the .out file? Are they syncing it back to the submit directory as the job is running?

[sidharath]

Originally Posted by singer1812

You would need to talk to your HPC people on this. I have no idea how they have it setup.

How are you able to see the .out file? Are they syncing it back to the submit directory as the job is running?

I am using a secure shell kind of software (winSCP) to transfer the files (def files and initial results) to the server and then executing the run on the server.
The .out file is being updated on real time in the run directory over the server and I am seeing the results from there.
It's more like working on a remote computer/server.
I'll talk to them if somehow we can access the "home" in server from the local machine directly but I highly doubt that since they use a private ssh key just to connect with the server.

Thank you