How to build solver for HPC cluster and submit simulation project to the queue under Windows OS

MFiX
#1

Dear All,

I have the possibility to have access to HPC resources,

  1. I want to know whether there are specific requirements on the type or configurations of the HPC cluster MFiX can work on.

  2. How to build the solver for HPC cluster and submit simulation project to the queue under Windows OS?

.

  1. How to retrieve the data from the HPC cluster? What if the data is more than 100GB? What else special things to pay attention to?

Thank you.

#2

  1. If you can install Anaconda on the cluster, you can install MFiX

  2. You have to do the compilation on the HPC cluster. You cannot build the solver under Windows and run it on a Linux cluster (I doubt your cluster is running Windows…).

There are options in the “Run” popup for using a script to submit your job to a queue manager (Torque, slurm, etc)

  1. With the current implementation of MFiX, the GUI and the solver rely on a shared filesystem. This filesystem contains the solver executable and all of its input and output files.

The GUI simply reads files from the project directory, so the project directory must be shared between the compute nodes and the host where you are running the GUI. In practice this means that people do not run the GUI on their own laptop, but on one of the cluster nodes, using remote display to view the GUI. It might be possible to run the GUI locally but you would need to set up a shared filesystem. You also need network connectivity so that the GUI can monitor the progress of the solver and issue start/pause/stop commands, this is done via HTTP. Generally, your compute nodes will be behind a firewall without network access to the outside, so this might require setting up tunneling.

Shared filesystem and network tunneling are beyond the scope of what we can support on the forum, but you might be able to make it work. It’s probably just easier to run the GUI on the cluster, like most folks do.

– Charles

#3

Hi Charles, thanks for quick reply. Considering it is slow to run the animation directly for computationally heavy project, I do not know whether HPC cluster can speed up the playing speed of animation or not…, I am considering 1. First run GUI with SMS mode to check generation of correct mesh with my PC, 2. using command line to run the project on cluster and extract the data to the local and process data with “paraview” or python notebook, finally generate animation on the local.

Do you think the above steps are possible? And regarding building solver for HPC cluster, are there any differences from doing that on PC? Can you provide me related tutorials? Thank you.

#4

Building the solver is covered in the manual https://mfix.netl.doe.gov/doc/mfix/21.3/html/solver/build.html

1 Like