Altair PBS

Launch a Simulation via PBS Access

Figure 1.
  1. Browse to your PBS Access server and login.
  2. Click the Jobs menu and select Submit Job > nanoFluidX
  3. Select the nanoFluidX version to be used and enter a Job name.
  4. Select (or drag-and-drop) the input files to the browser so that they are uploaded.
    Note: It is possible to upload a *.zip file when there are many files or if the file size is large. This will be automatically extracted at runtime.
  5. Select the number of GPUs to use and click Submit Job.

Test PBS

When Altair PBS is used as the scheduler on the target system, a test run of nanoFluidX could be carried out through PBS. For further details on setting up PBS for use with GPU resources, refer to the relevant PBS documentation.
Note: The exact steps may change depending on the PBS settings on your particular system.
  1. Log in as a regular user. Do not use root user.
  2. Run system tests without using PBS, following steps 1 through 4 outlined in the Single Node section.
  3. Copy dambreak case from demo_cases to the location of choice: 
    <LOC>: cp <INSTALL_DIR>/altair/hwcfdsolvers/nanofluidx/linux64/demo_cases/dambreak <LOC>/dambreak
    1. For multi node systems with shared storage: 
      cp <INSTALL_DIR>/altair/hwcfdsolvers/nanofluidx/linux64/demo_cases/dambreak <LOC>/dambreak
    2. Repeat the copy operation for all nodes of a multi node system with local storage.
  4. Prepare the environment in your PBS script by adding one of the following:
    • Single node:
      nfxmpirun_pbssinglenode -np 1 $nFX_SP -i dambreak_3D_H300_dx0p005_wt.cfg -m 10
    • Multi node:
      <INSTALL_DIR>/altair/hwcfdsolvers/nanofluidx/linux64/set_nFX_environment.sh multinode
    1. PBS must produce a HOSTFILE where each node is repeated once per GPU on each node.
    2. Set X equal to the number of GPUs across all nodes.
  5. Launch nanoFluidX in your PBS script by adding the following:
    • Single node:
      nfxmpirun_pbssinglenode -np 1 $nFX_SP -i dambreak_3D_H300_dx0p005_wt.cfg -m 10
    • Multi node:
      nfxmpirun_pbsmultinode -np X $nFX_SP -i dambreak_3D_H300_dx0p005_wt.cfg -m 10
    1. PBS must produce a HOSTFILE where each node is repeated once per GPU on each node.
    2. Set X equal to the number of GPUs across all nodes.
  6. The test run is successful when the message Finished (reached maximum steps) appears at the end of the PBS output file.
    Note: Simulation time may vary.
    Figure 2.